This document summarizes the seam carving method for content-aware image resizing. Seam carving allows images to be resized while preserving important image content. A seam is defined as an optimal 8-connected path of pixels from top to bottom or left to right of an image. By repeatedly carving out or inserting seams, the aspect ratio of an image can be changed. This provides an alternative to standard scaling or cropping methods. The document outlines the seam carving algorithm and energy functions used to define important image content. Applications discussed include aspect ratio change, image retargeting, enhancement and object removal.
This presentation is based on the SIGGRAPH 2016 paper which published on research gate. I just prepared and studied the algorithm. https://www.researchgate.net/publication/305217688_Mapping_virtual_and_physical_reality
Digital Ortho Image Creation of Hall County Aerial Photos Papermpadams77
Special Topics Project Paper “Digital Ortho Image Creation of Hall County Aerial Photos” which I presented at the Florida Academy of Science and Georgia Academy of Science Joint Conference held in Jacksonville, FL March 14th and 15th of 2008
Presentation on Digital Image ProcessingSalim Hosen
Digital image processing is the use of a digital computer to process digital images through an algorithm. As a subcategory or field of digital signal processing, digital image processing has many advantages over analog image processing.
This presentation is based on the SIGGRAPH 2016 paper which published on research gate. I just prepared and studied the algorithm. https://www.researchgate.net/publication/305217688_Mapping_virtual_and_physical_reality
Digital Ortho Image Creation of Hall County Aerial Photos Papermpadams77
Special Topics Project Paper “Digital Ortho Image Creation of Hall County Aerial Photos” which I presented at the Florida Academy of Science and Georgia Academy of Science Joint Conference held in Jacksonville, FL March 14th and 15th of 2008
Presentation on Digital Image ProcessingSalim Hosen
Digital image processing is the use of a digital computer to process digital images through an algorithm. As a subcategory or field of digital signal processing, digital image processing has many advantages over analog image processing.
In computer science, digital image processing is the use of computer algorithms to perform image processing on digital images. As a subcategory or field of digital signal processing, digital image processing has many advantages over analog image processing.It allows a much wider range of algorithms to be applied to the input data and can avoid problems such as the build-up of noise and signal ...
digital image processing pdf
digital image processing books
digital image processing textbook pdf
digital image processing textbook
digital image processing pdf book
digital image processing gonzalez pdf
digital image processing 4th pdf
digital image processing 3rd pdf
digital image processing slides
history of image processing
digital image processing third edition
digital image processing pdf
digital image processing gonzalez ppt
digital image processing 3rd edition pdf
digital image processing third edition pdf
image processing basics
Depth of Field Image Segmentation Using Saliency Map and Energy Mapping Techn...ijsrd.com
Image plays a vital role in image processing. In Image processing Depth of Field is to segment the relevant object from an Image. Depth of Field is the space between the near and extreme objects in a scene. The objective of this work is to segment the image using Low Depth of Field .Unsupervised segmentation is used to find low depth of field image. Saliency map and curve evaluation method is created and initialized for the image. Energy map have been employed so as to bring the desired result. Lipschitz function is used to generate the mathematical view of representation. Various Iteration methods have shown the graphical representation of an image. The Segmented results have shown the Object detection in an image.
MAGE Q UALITY A SSESSMENT OF T ONE M APPED I MAGESijcga
This paper proposes an objective assessment method
for perceptual image quality of tone mapped images.
Tone mapping algorithms are used to display high dy
namic range (HDR) images onto standard display
devices that have low dynamic range (LDR). The prop
osed method implements visual attention to define
perceived structural distortion regions in LDR imag
es, so that a reasonable measurement of distortion
between HDR and LDR images can be performed. Since
the human visual system is sensitive to structural
information, quality metrics that can measure struc
tural similarity between HDR and LDR images are
used. Experimental results with a number of HDR and
tone mapped image pairs show the effectiveness of
the proposed method.
Image Quality Assessment of Tone Mapped Images ijcga
This paper proposes an objective assessment method for perceptual image quality of tone mapped images. Tone mapping algorithms are used to display high dynamic range (HDR) images onto standard display devices that have low dynamic range (LDR). The proposed method implements visual attention to define perceived structural distortion regions in LDR images, so that a reasonable measurement of distortion between HDR and LDR images can be performed. Since the human visual system is sensitive to structural information, quality metrics that can measure structural similarity between HDR and LDR images are used. Experimental results with a number of HDR and tone mapped image pairs show the effectiveness of the proposed method.
Reversible Enhancement of Reversibly Data Embedded Images using HVS Character...CSCJournals
In the recent, HVS based reversible data embedding is emerging. In a reversible data embedding scheme, in the process of achieving reversibility and embedding the data, it induces some distortions into marked image/video. If the distortions in marked image are random due to reversible embedding, it may not reveal much information about the existence of hidden data; on the contrary, if the distortions are repetitive geometric patterns which arises when an embedding scheme's emphasis is on capacity and reversibility, then the marked media with geometric distortions may reveal the existence of the hidden data. Further, in applications involving satellite images, medical, military, fine arts images, etc. which involves reversible embedding, it requires to show the details for inspection. Due to repetitive geometric distortions as in [6] and [7], the visual quality of image degrades. Hence, there is a need for enhancing the reversibly embedded images. To tackle this issue, if any exiting image enhancement scheme is used, it incur in loss of hidden data as well as the restoration of image to original form is not possible; hence, violating the requirement of reversibility for such applications. To address this problem we propose a reversible image enhancement scheme for reversibly data embedded images by using HVS characteristics of DCT. The experimental results show that the visual quality has been improved in terms of PSNR, PSNR-HVS, PSNR-HVS-M, and MSSIM when compared to S. Gujjunoori et al. scheme [6] and S. Gujjunoori et al. scheme [7], which contains repetitive geometrical distortions in marked media.
High Performance Computing for Satellite Image Processing and Analyzing – A ...Editor IJCATR
High Performance Computing (HPC) is the recently developed technology in the field of computer science, which evolved
due to meet increasing demands for processing speed and analysing/processing huge size of data sets. HPC brings together several
technologies such as computer architecture, algorithm, programs and system software under one canopy to solve/handle advanced
complex problems quickly and effectively. It is a crucial element today to gather and process large amount of satellite (remote sensing)
data which is the need of an hour. In this paper, we review recent development in HPC technology (Parallel, Distributed and Cluster
Computing) for satellite data processing and analysing. We attempt to discuss the fundamentals of High Performance Computing
(HPC) for satellite data processing and analysing, in a way which is easy to understand without much previous background. We sketch
the various HPC approach such as Parallel, Distributed & Cluster Computing and subsequent satellite data processing & analysing
methods like geo-referencing, image mosaicking, image classification, image fusion and Morphological/neural approach for hyperspectral satellite data. Collective, these works deliver a snapshot, tables and algorithms of the recent developments in those sectors and
offer a thoughtful perspective of the potential and promising challenges of satellite data processing and analysing using HPC
paradigms.
In computer science, digital image processing is the use of computer algorithms to perform image processing on digital images. As a subcategory or field of digital signal processing, digital image processing has many advantages over analog image processing.It allows a much wider range of algorithms to be applied to the input data and can avoid problems such as the build-up of noise and signal ...
digital image processing pdf
digital image processing books
digital image processing textbook pdf
digital image processing textbook
digital image processing pdf book
digital image processing gonzalez pdf
digital image processing 4th pdf
digital image processing 3rd pdf
digital image processing slides
history of image processing
digital image processing third edition
digital image processing pdf
digital image processing gonzalez ppt
digital image processing 3rd edition pdf
digital image processing third edition pdf
image processing basics
Depth of Field Image Segmentation Using Saliency Map and Energy Mapping Techn...ijsrd.com
Image plays a vital role in image processing. In Image processing Depth of Field is to segment the relevant object from an Image. Depth of Field is the space between the near and extreme objects in a scene. The objective of this work is to segment the image using Low Depth of Field .Unsupervised segmentation is used to find low depth of field image. Saliency map and curve evaluation method is created and initialized for the image. Energy map have been employed so as to bring the desired result. Lipschitz function is used to generate the mathematical view of representation. Various Iteration methods have shown the graphical representation of an image. The Segmented results have shown the Object detection in an image.
MAGE Q UALITY A SSESSMENT OF T ONE M APPED I MAGESijcga
This paper proposes an objective assessment method
for perceptual image quality of tone mapped images.
Tone mapping algorithms are used to display high dy
namic range (HDR) images onto standard display
devices that have low dynamic range (LDR). The prop
osed method implements visual attention to define
perceived structural distortion regions in LDR imag
es, so that a reasonable measurement of distortion
between HDR and LDR images can be performed. Since
the human visual system is sensitive to structural
information, quality metrics that can measure struc
tural similarity between HDR and LDR images are
used. Experimental results with a number of HDR and
tone mapped image pairs show the effectiveness of
the proposed method.
Image Quality Assessment of Tone Mapped Images ijcga
This paper proposes an objective assessment method for perceptual image quality of tone mapped images. Tone mapping algorithms are used to display high dynamic range (HDR) images onto standard display devices that have low dynamic range (LDR). The proposed method implements visual attention to define perceived structural distortion regions in LDR images, so that a reasonable measurement of distortion between HDR and LDR images can be performed. Since the human visual system is sensitive to structural information, quality metrics that can measure structural similarity between HDR and LDR images are used. Experimental results with a number of HDR and tone mapped image pairs show the effectiveness of the proposed method.
Reversible Enhancement of Reversibly Data Embedded Images using HVS Character...CSCJournals
In the recent, HVS based reversible data embedding is emerging. In a reversible data embedding scheme, in the process of achieving reversibility and embedding the data, it induces some distortions into marked image/video. If the distortions in marked image are random due to reversible embedding, it may not reveal much information about the existence of hidden data; on the contrary, if the distortions are repetitive geometric patterns which arises when an embedding scheme's emphasis is on capacity and reversibility, then the marked media with geometric distortions may reveal the existence of the hidden data. Further, in applications involving satellite images, medical, military, fine arts images, etc. which involves reversible embedding, it requires to show the details for inspection. Due to repetitive geometric distortions as in [6] and [7], the visual quality of image degrades. Hence, there is a need for enhancing the reversibly embedded images. To tackle this issue, if any exiting image enhancement scheme is used, it incur in loss of hidden data as well as the restoration of image to original form is not possible; hence, violating the requirement of reversibility for such applications. To address this problem we propose a reversible image enhancement scheme for reversibly data embedded images by using HVS characteristics of DCT. The experimental results show that the visual quality has been improved in terms of PSNR, PSNR-HVS, PSNR-HVS-M, and MSSIM when compared to S. Gujjunoori et al. scheme [6] and S. Gujjunoori et al. scheme [7], which contains repetitive geometrical distortions in marked media.
High Performance Computing for Satellite Image Processing and Analyzing – A ...Editor IJCATR
High Performance Computing (HPC) is the recently developed technology in the field of computer science, which evolved
due to meet increasing demands for processing speed and analysing/processing huge size of data sets. HPC brings together several
technologies such as computer architecture, algorithm, programs and system software under one canopy to solve/handle advanced
complex problems quickly and effectively. It is a crucial element today to gather and process large amount of satellite (remote sensing)
data which is the need of an hour. In this paper, we review recent development in HPC technology (Parallel, Distributed and Cluster
Computing) for satellite data processing and analysing. We attempt to discuss the fundamentals of High Performance Computing
(HPC) for satellite data processing and analysing, in a way which is easy to understand without much previous background. We sketch
the various HPC approach such as Parallel, Distributed & Cluster Computing and subsequent satellite data processing & analysing
methods like geo-referencing, image mosaicking, image classification, image fusion and Morphological/neural approach for hyperspectral satellite data. Collective, these works deliver a snapshot, tables and algorithms of the recent developments in those sectors and
offer a thoughtful perspective of the potential and promising challenges of satellite data processing and analysing using HPC
paradigms.
http://inarocket.com
Learn BEM fundamentals as fast as possible. What is BEM (Block, element, modifier), BEM syntax, how it works with a real example, etc.
How to Build a Dynamic Social Media PlanPost Planner
Stop guessing and wasting your time on networks and strategies that don’t work!
Join Rebekah Radice and Katie Lance to learn how to optimize your social networks, the best kept secrets for hot content, top time management tools, and much more!
Watch the replay here: bit.ly/socialmedia-plan
Lightning Talk #9: How UX and Data Storytelling Can Shape Policy by Mika Aldabaux singapore
How can we take UX and Data Storytelling out of the tech context and use them to change the way government behaves?
Showcasing the truth is the highest goal of data storytelling. Because the design of a chart can affect the interpretation of data in a major way, one must wield visual tools with care and deliberation. Using quantitative facts to evoke an emotional response is best achieved with the combination of UX and data storytelling.
Content personalisation is becoming more prevalent. A site, it's content and/or it's products, change dynamically according to the specific needs of the user. SEO needs to ensure we do not fall behind of this trend.
Succession “Losers”: What Happens to Executives Passed Over for the CEO Job?
By David F. Larcker, Stephen A. Miles, and Brian Tayan
Stanford Closer Look Series
Overview:
Shareholders pay considerable attention to the choice of executive selected as the new CEO whenever a change in leadership takes place. However, without an inside look at the leading candidates to assume the CEO role, it is difficult for shareholders to tell whether the board has made the correct choice. In this Closer Look, we examine CEO succession events among the largest 100 companies over a ten-year period to determine what happens to the executives who were not selected (i.e., the “succession losers”) and how they perform relative to those who were selected (the “succession winners”).
We ask:
• Are the executives selected for the CEO role really better than those passed over?
• What are the implications for understanding the labor market for executive talent?
• Are differences in performance due to operating conditions or quality of available talent?
• Are boards better at identifying CEO talent than other research generally suggests?
Our life’s important part is Image. Without disturbing its overall structure of images, we can
remove the unwanted part of image with the help of image inpainting. There is simpler the inpainting of
the low resolution images than that of the high resolution images. In this system low resolution image
contained in different super resolution image inpainting methodologies and there are combined all these
methodologies to form the highly in painted image results. For this reason our system uses the super
resolution algorithm which is responsiblefor inpainting of singleimage.
Image Segmentation Using Pairwise Correlation ClusteringIJERA Editor
A pairwise hypergraph based image segmentation framework is formulated in a supervised manner for various images. The image segmentation is to infer the edge label over the pairwise hypergraph by maximizing the normalized cuts. Correlation clustering which is a graph partitioning algorithm, was shown to be effective in a number of applications such as identification, clustering of documents and image segmentation.The partitioning result is derived from a algorithm to partition a pairwise graph into disjoint groups of coherent nodes. In the pairwise correlation clustering, the pairwise graph which is used in the correlation clustering is generalized to a superpixel graph where a node corresponds to a superpixel and a link between adjacent superpixels corresponds to an edge. This pairwise correlation clustering also considers the feature vector which extracts several visual cues from a superpixel, including brightness, color, texture, and shape. Significant progress in clustering has been achieved by algorithms that are based on pairwise affinities between the datasets. The experimental results are shown by calculating the typical cut and inference in an undirected graphical model and datasets.
Comparative Study and Analysis of Image Inpainting TechniquesIOSR Journals
Abstract: Image inpainting is a technique to fill missing region or reconstruct damage area from an image.It
removes an undesirable object from an image in visually plausible way.For filling the part of image, it use
information from the neighboring area. In this dissertation work, we present a Examplar based method for
filling in the missing information in an image, which takes structure synthesis and texture sysnthesis together.
In exemplar based approach it used local information from an image to patch propagation.We have also
implement Nonlocal Mean approach for exemplar based image inpainting.In Nonlocal mean approach it find
multiple samples of best exemplar patches for patch propagation and weight their contribution according to
their similarity to the neighborhood under evaluation. We have further extended this algorithm by considering
collaborative filtering method to synthesize and propagate with multiple samples of best exemplar patches. We
have to preformed experiment on many images and found that our algorithm successfully inpaint the target
region.We have tested the accuracy of our algorithm by finding parameter like PSNR and compared PSNR
value for all three different approaches.
Keywords: Texture Synthesis, Structure Synthesis, Patch Propagation ,imageinpainting ,nonlocal approach,
collabrative filtering.
To Get any Project for CSE, IT ECE, EEE Contact Me @ 09666155510, 09849539085 or mail us - ieeefinalsemprojects@gmail.com-Visit Our Website: www.finalyearprojects.org
Presentation on deformable model for medical image segmentationSubhash Basistha
Introduction to Image Processing
Steps of Image Processing
Types of Image Processing
Introduction to Image Segmentation
Introduction to Medical Image Segmentation
Application of Image Segmentation
Example of Image Segmentation
Need for Deformable Model
What is Deformable Model??
Types of Deformable Model
A Detailed Analysis on Feature Extraction Techniques of Panoramic Image Stitc...IJEACS
Image stitching is a technique which is used for attaining a high resolution panoramic image. In this technique, distinct aesthetic images that are imaged from different view and angles are combined together to produce a panoramic image. In the field of computer graphics, photographic and computer vision, Image stitching techniques are considered as current research areas. For obtaining a stitched image it becomes mandatory that one should have the knowledge of geometric relations among multiple image co-ordinate system [1].First, image stitching will be done based on feature key point matches. Final image with seam will be blended with image blending technique. Hence in this paper we are going to address multiple distinct techniques like some invariant features as Scale Invariant Feature Transform and Speeded up Robust Transform and Corner techniques as Harris Corner Detection Technique that are useful in sorting out the issues related with stitching of images.
Adaptive Image Resizing using Edge Contrastingijtsrd
Zooming is an important image processing operation. It can be termed as the process of enlarging or magnifying the image to a given factor. Indiscriminate application of a function to an image in order to resample it, produces aliasing, edge blurring. So the objective is to reduce these artifacts.This paper considers distinctive interpolation systems identified with versatile techniques with innate abilities to ensure sharp edges and subtleties. It is a versatile resampling calculation for zooming up pictures In this work, a versatile edge improvement procedure is proposed for two dimensional 2 D picture scaling application. The foreseen picture scaling calculation comprises of an edge identifier, bilinear interpolation and Sobel filter. The bilinear interpolation characterizes the power of the scaled pixel with the weighted normal of the four neighboring pixels, inserted pictures become smooth and loss of edge data. The versatile edge upgrade procedure is utilized to secure the edge includes successfully, to accomplish better picture quality and to dodge the edge data. The Sobel filter endeavors to decrease the commotion, in obscured and mutilated edges which is delivered by bilinear interpolation. A mathematical control and equipment sharing strategy are utilized to lessen registering asset of the bilinear interpolation. The examination shows that edges are very much safeguarded and interpolation artifacts obscuring, jaggies are decreased To contrast existing algorithms and proposed strategy calculation, we have taken original pictures and results for discussion. And we have gone to the choice that proposed calculation is superior to the current algorithms. We have looked at the images by two different ways – Mean Square Error MSE and Peak Signal to Noise Ratio PSNR . Mohd Sadiq Abdul Aziz | Dr. Bharti Chourasia "Adaptive Image Resizing using Edge Contrasting" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-6 , October 2020, URL: https://www.ijtsrd.com/papers/ijtsrd35789.pdf Paper Url: https://www.ijtsrd.com/engineering/electronics-and-communication-engineering/35789/adaptive-image-resizing-using-edge-contrasting/mohd-sadiq-abdul-aziz
CASAIR: CONTENT AND SHAPE-AWARE IMAGE RETARGETING AND ITS APPLICATIONSNexgen Technology
TO GET THIS PROJECT COMPLETE SOURCE ON SUPPORT WITH EXECUTION PLEASE CALL BELOW CONTACT DETAILS
MOBILE: 9791938249, 0413-2211159, WEB: WWW.NEXGENPROJECT.COM,WWW.FINALYEAR-IEEEPROJECTS.COM, EMAIL:Praveen@nexgenproject.com
NEXGEN TECHNOLOGY provides total software solutions to its customers. Apsys works closely with the customers to identify their business processes for computerization and help them implement state-of-the-art solutions. By identifying and enhancing their processes through information technology solutions. NEXGEN TECHNOLOGY help it customers optimally use their resources.
Removal of Unwanted Objects using Image Inpainting - a Technical ReviewIJERA Editor
Image In painting, the technique to change image in undetectable structure, it itself is an ancient art. There are
various goals and applications of image in painting which includes restoration of damaged painting and also to
replace/remove the selected objects. This paper, describes various techniques that can help in removing
unwanted objects from image. Even the in painting fundamentals are directly further, most inpainting techniques
available in the literature are difficult to understand and implement.
Encryption in Microsoft 365 - ExpertsLive Netherlands 2024Albert Hoitingh
In this session I delve into the encryption technology used in Microsoft 365 and Microsoft Purview. Including the concepts of Customer Key and Double Key Encryption.
Pushing the limits of ePRTC: 100ns holdover for 100 daysAdtran
At WSTS 2024, Alon Stern explored the topic of parametric holdover and explained how recent research findings can be implemented in real-world PNT networks to achieve 100 nanoseconds of accuracy for up to 100 days.
Why You Should Replace Windows 11 with Nitrux Linux 3.5.0 for enhanced perfor...SOFTTECHHUB
The choice of an operating system plays a pivotal role in shaping our computing experience. For decades, Microsoft's Windows has dominated the market, offering a familiar and widely adopted platform for personal and professional use. However, as technological advancements continue to push the boundaries of innovation, alternative operating systems have emerged, challenging the status quo and offering users a fresh perspective on computing.
One such alternative that has garnered significant attention and acclaim is Nitrux Linux 3.5.0, a sleek, powerful, and user-friendly Linux distribution that promises to redefine the way we interact with our devices. With its focus on performance, security, and customization, Nitrux Linux presents a compelling case for those seeking to break free from the constraints of proprietary software and embrace the freedom and flexibility of open-source computing.
DevOps and Testing slides at DASA ConnectKari Kakkonen
My and Rik Marselis slides at 30.5.2024 DASA Connect conference. We discuss about what is testing, then what is agile testing and finally what is Testing in DevOps. Finally we had lovely workshop with the participants trying to find out different ways to think about quality and testing in different parts of the DevOps infinity loop.
Epistemic Interaction - tuning interfaces to provide information for AI supportAlan Dix
Paper presented at SYNERGY workshop at AVI 2024, Genoa, Italy. 3rd June 2024
https://alandix.com/academic/papers/synergy2024-epistemic/
As machine learning integrates deeper into human-computer interactions, the concept of epistemic interaction emerges, aiming to refine these interactions to enhance system adaptability. This approach encourages minor, intentional adjustments in user behaviour to enrich the data available for system learning. This paper introduces epistemic interaction within the context of human-system communication, illustrating how deliberate interaction design can improve system understanding and adaptation. Through concrete examples, we demonstrate the potential of epistemic interaction to significantly advance human-computer interaction by leveraging intuitive human communication strategies to inform system design and functionality, offering a novel pathway for enriching user-system engagements.
Transcript: Selling digital books in 2024: Insights from industry leaders - T...BookNet Canada
The publishing industry has been selling digital audiobooks and ebooks for over a decade and has found its groove. What’s changed? What has stayed the same? Where do we go from here? Join a group of leading sales peers from across the industry for a conversation about the lessons learned since the popularization of digital books, best practices, digital book supply chain management, and more.
Link to video recording: https://bnctechforum.ca/sessions/selling-digital-books-in-2024-insights-from-industry-leaders/
Presented by BookNet Canada on May 28, 2024, with support from the Department of Canadian Heritage.
Securing your Kubernetes cluster_ a step-by-step guide to success !KatiaHIMEUR1
Today, after several years of existence, an extremely active community and an ultra-dynamic ecosystem, Kubernetes has established itself as the de facto standard in container orchestration. Thanks to a wide range of managed services, it has never been so easy to set up a ready-to-use Kubernetes cluster.
However, this ease of use means that the subject of security in Kubernetes is often left for later, or even neglected. This exposes companies to significant risks.
In this talk, I'll show you step-by-step how to secure your Kubernetes cluster for greater peace of mind and reliability.
GraphRAG is All You need? LLM & Knowledge GraphGuy Korland
Guy Korland, CEO and Co-founder of FalkorDB, will review two articles on the integration of language models with knowledge graphs.
1. Unifying Large Language Models and Knowledge Graphs: A Roadmap.
https://arxiv.org/abs/2306.08302
2. Microsoft Research's GraphRAG paper and a review paper on various uses of knowledge graphs:
https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/
Le nuove frontiere dell'AI nell'RPA con UiPath Autopilot™UiPathCommunity
In questo evento online gratuito, organizzato dalla Community Italiana di UiPath, potrai esplorare le nuove funzionalità di Autopilot, il tool che integra l'Intelligenza Artificiale nei processi di sviluppo e utilizzo delle Automazioni.
📕 Vedremo insieme alcuni esempi dell'utilizzo di Autopilot in diversi tool della Suite UiPath:
Autopilot per Studio Web
Autopilot per Studio
Autopilot per Apps
Clipboard AI
GenAI applicata alla Document Understanding
👨🏫👨💻 Speakers:
Stefano Negro, UiPath MVPx3, RPA Tech Lead @ BSP Consultant
Flavio Martinelli, UiPath MVP 2023, Technical Account Manager @UiPath
Andrei Tasca, RPA Solutions Team Lead @NTT Data
The Metaverse and AI: how can decision-makers harness the Metaverse for their...Jen Stirrup
The Metaverse is popularized in science fiction, and now it is becoming closer to being a part of our daily lives through the use of social media and shopping companies. How can businesses survive in a world where Artificial Intelligence is becoming the present as well as the future of technology, and how does the Metaverse fit into business strategy when futurist ideas are developing into reality at accelerated rates? How do we do this when our data isn't up to scratch? How can we move towards success with our data so we are set up for the Metaverse when it arrives?
How can you help your company evolve, adapt, and succeed using Artificial Intelligence and the Metaverse to stay ahead of the competition? What are the potential issues, complications, and benefits that these technologies could bring to us and our organizations? In this session, Jen Stirrup will explain how to start thinking about these technologies as an organisation.
Removing Uninteresting Bytes in Software FuzzingAftab Hussain
Imagine a world where software fuzzing, the process of mutating bytes in test seeds to uncover hidden and erroneous program behaviors, becomes faster and more effective. A lot depends on the initial seeds, which can significantly dictate the trajectory of a fuzzing campaign, particularly in terms of how long it takes to uncover interesting behaviour in your code. We introduce DIAR, a technique designed to speedup fuzzing campaigns by pinpointing and eliminating those uninteresting bytes in the seeds. Picture this: instead of wasting valuable resources on meaningless mutations in large, bloated seeds, DIAR removes the unnecessary bytes, streamlining the entire process.
In this work, we equipped AFL, a popular fuzzer, with DIAR and examined two critical Linux libraries -- Libxml's xmllint, a tool for parsing xml documents, and Binutil's readelf, an essential debugging and security analysis command-line tool used to display detailed information about ELF (Executable and Linkable Format). Our preliminary results show that AFL+DIAR does not only discover new paths more quickly but also achieves higher coverage overall. This work thus showcases how starting with lean and optimized seeds can lead to faster, more comprehensive fuzzing campaigns -- and DIAR helps you find such seeds.
- These are slides of the talk given at IEEE International Conference on Software Testing Verification and Validation Workshop, ICSTW 2022.
Dev Dives: Train smarter, not harder – active learning and UiPath LLMs for do...UiPathCommunity
💥 Speed, accuracy, and scaling – discover the superpowers of GenAI in action with UiPath Document Understanding and Communications Mining™:
See how to accelerate model training and optimize model performance with active learning
Learn about the latest enhancements to out-of-the-box document processing – with little to no training required
Get an exclusive demo of the new family of UiPath LLMs – GenAI models specialized for processing different types of documents and messages
This is a hands-on session specifically designed for automation developers and AI enthusiasts seeking to enhance their knowledge in leveraging the latest intelligent document processing capabilities offered by UiPath.
Speakers:
👨🏫 Andras Palfi, Senior Product Manager, UiPath
👩🏫 Lenka Dulovicova, Product Program Manager, UiPath
2. 10-2 • Avidan et al.
both vertical and horizontal directions we define multi-size images. by moving a slider. In their video retargeting work they use a com-
Such images can continuously change their size in a content-aware bination of image and saliency maps to find the ROI. Then they use
manner. A designer can author a multi-size image once, and the a combination of cropping, virtual pan and shot cuts to retarget the
client application, depending on the size needed, can resize the im- video frames.
age in real time to fit the exact layout or the display.
Seam carving can support several types of energy functions such as Setlur et al. [2005] proposed an automatic, non-photorealistic al-
gradient magnitude, entropy, visual saliency, eye-gaze movement, gorithm for retargetting large images to small size displays. This is
and more. The removal or insertion processes are parameter free; done by decomposing the image into a background layer and fore-
however, to allow interactive control, we also provide a scribble- ground objects. The retargeting algorithm segments an image into
based user interface for adding weights to the energy of an image regions, identifies important regions, removes them, fills the result-
and guide the desired results. This tool can also be used for author- ing gaps, resize the remaining image, and re-insert the important
ing multi-size images. To summarize, our main contributions are as region.
follows:
The first solution to the general problem of warping an image into
• Define seam carving and present its properties. an arbitrary shape while preserving user-specified features was re-
cently proposed by Gal et al. [2006]. The feature-aware warping is
• Present algorithm for image enlarging using seam insertions.
achieved by a particular formulation of the Laplacian editing tech-
• Use seams for content-aware image size manipulations. nique, suited to accommodate similarity constraints on parts of the
domain. Since local constraints are propagated by the global opti-
• Define multi-size images for continuous image retargeting. mization process, not all the constraints can always be satisfied at
once. Our algorithm is discrete, so carving a single seam has no
affect on the rest of the image.
2 Background
Image resizing is a standard tool in many image processing appli- The use of seams for image editing is prevalent. Agarwala et al.
cations. It works by uniformly resizing the image to a target size. [2004] describe an interactive Digital Photomontage system that
Recently, there is a growing interest in image retargeting that seeks finds perfect seams to combine parts of a set of photographs into
to change the size of the image while maintaining the important fea- a single composite picture, using minimal user assistance. Jia et al.
tures intact, where these features can be either detected top-down [2006] proposed Drag-and-Drop Pasting that extends the Poisson
or bottom-up. Top down methods use tools such as face detectors Image Editing technique [Perez et al. 2003] to compute an optimal
[Viola and Jones 2001] to detect important regions in the image, boundary (i.e. seam) between the source and target images. Rother
whereas bottom-up methods rely on visual saliency methods [Itti et al. [2006] developed AutoCollage, a program that automatically
et al. 1999] to construct a visual saliency map of the image. Once creates a collage image from a collection of images. This process
the saliency map is constructed, cropping can be used to display requires, among other things, finding optimal boundaries, or seams,
the most important region of the image. Suh et al. [2003] pro- between many image fragments. None of the above methods dis-
posed automatic thumbnail creation, based on either a saliency map cuss the problem of image retargeting. A notable exception is the
or the output of a face detector. The large image is then cropped work of Wang and Cohen [2006] that proposes to simultaneously
to capture the most salient region in the image. Similarly, Chen et solve matting and compositing. They allow the user to scale the
al. [2003] considered the problem of adapting images to mobile size of the foreground object and paste it back on the original back-
devices. In their approach the most important region in the image ground. Zomet et al. [2005] evaluated several cost functions for
is automatically detected and transmitted to the mobile device. Liu seamless image stitching and concluded that minimizing an l1 error
et al. [2003] also addressed image retargeting to mobile devices, norm between the gradients of the stitched image and the gradi-
suggesting to trade time for space. Given a collection of regions of ents of the input images performed well in general. Computing the
interest, they construct an optimal path through these regions and seam can be done in a variety of ways, including Dijkstra’s shortest
display them serially, one after the other, to the user. Santella et path algorithm [1998], dynamic programming [2001] or graph cuts
al. [2006] use eye tracking, in addition to composition rules to crop [2001].
images intelligently. All these methods achieve impressive results,
but rely on traditional image resizing and cropping operations to Changing the size of the image has been extensively studied in
actually change the size of the image. the field of texture synthesis, where the goal is to generate a large
texture image from a small one. Efros et al. [2001] find seams
Jacobs et al. [2003] consider an adaptive grid-based document lay- that minimize the error surface defined by two overlapping texture
out system that maintains a clear separation between content and patches. This way, the original small texture image is quilted to
template. The page designer constructs several possible templates form a much larger texture image. This was later extended to han-
and when the content is displayed the most suitable template is dle both image and video texture synthesis by Kwatra et al. [2003]
used. The templates can use different discrete alternatives of an that showed how to increase the space and time dimensions of the
image if they are provided, but no specific reference to image resiz- original texture video.
ing is made.
A compromise between image resizing and image cropping is to As for object removal, Bertalmio et al. [2000] proposed an im-
introduce a non-linear, data dependent scaling. Such a method was age inpainting method that smoothly propagates information from
proposed by Liu and Gleicher [2005; 2006] for image and video the boundaries inwards, simulating techniques used by professional
retargeting. For image retargeting they find the Region-Of-Interest restorators. Patch based approaches [Drori et al. 2003; Criminisi
(ROI) and construct a novel Fisheye-View warp that essentially ap- et al. 2003; Bertalmio et al. 2003] use automatic guidance to deter-
plies a piecewise linear scaling function in each dimension to the mine synthesis ordering, which considerably improves the quality
image. This way the ROI is maintained while the rest of the image of the results. And recently, Sun et al. [2005] proposed an inter-
is warped. The retargeting can be done in interactive rates, once the active method to handle inpainting in case of missing strong visual
ROI is found, so the user can control the desired size of the image structure, by propagating structure along used-specified curves.
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
3. Seam Carving for Content-Aware Image Resizing • 10-3
(a) Original (b) Crop (c) Column (d) Seam (e) Pixel (f) Optimal
Figure 2: Results of 5 different strategies for reducing the width of an image. (a) the original image and its e1 energy function, (b) best
cropping, (c) removing columns with minimal energy, (d) seam removal, (e) removal of the pixel with the least amount of energy in each row,
and finally, (f) global removal of pixels with the lowest energy, regardless of their position. Figure 3 shows the energy preservation of each
strategy.
3 The Operator
Our approach to content-aware resizing is to remove pixels in a ju-
dicious manner. Therefor, the question is how to chose the pixels to
be removed? Intuitively, our goal is to remove unnoticeable pixels
that blend with their surroundings. This leads to the following sim-
ple energy function that was used in many figures in this paper such
as Figures 1, 6, 5, 8, 11, 12, 13 (we explore other energy functions
in subsection 3.2):
∂ ∂
e1 (I) = | I| + | I| (1)
∂x ∂y
Given an energy function, assume we need to reduce the image
width. One can think of several strategies to achieve this. For in-
stance, an optimal strategy to preserve energy (i.e., keep pixels with Figure 3: Image energy preservation. A comparison of the preser-
high energy value) would be to remove the pixels with lowest en- vation of content measured by the average pixel energy using five
ergy in ascending order. This destroys the rectangular shape of the different strategies of resizing. The actual images can be seen in
image, because we may remove a different number of pixels from Figure 2.
each row (see Figure 2(f)). If we want to prevent the image from
breaking we can remove an equal number of low energy pixels from
every row. This preserves the rectangular shape of the image but de- The visual impact is noticeable only along the path of the seam,
stroys the image content by creating a zigzag effect (Figure 2(e)). leaving the rest of the image intact. Note also that one can replace
To preserve both the shape and the visual coherence of the image the constraint |x(i) − x(i − 1)| ≤ 1 with |x(i) − x(i − 1)| ≤ k, and get
we can use Auto-cropping. That is, look for a sub-window, the size either a simple column (or row) for k = 0, a piecewise connected or
of the target image, that contains the highest energy (Figure 2(b)). even completely disconnected set of pixels for any value 1 ≤ k ≤ m.
Another possible strategy somewhat between removing pixels and
cropping is to remove whole columns with the lowest energy. Still, Given an energy function e, we can define the cost of a seam as
artifacts might appear in the resulting image (Figure 2(c)). There- E(s) = E(Is ) = ∑n e(I(si )). We look for the optimal seam s∗ that
i=1
fore, we need a resizing operator that will be less restrictive than minimizes this seam cost :
cropping or column removal, but can preserve the image content n
better than single pixel removals. This leads to our strategy of seam s∗ = min E(s) = min ∑ e(I(si )) (4)
carving (Figure 2(d)) and the definition of internal seams. s s
i=1
Formally, let I be an n × m image and define a vertical seam to be: The optimal seam can be found using dynamic programming. The
first step is to traverse the image from the second row to the last row
sx = {sx }n = {(x(i), i)}n , s.t. ∀i, |x(i) − x(i − 1)| ≤ 1,
i i=1 i=1 (2) and compute the cumulative minimum energy M for all possible
connected seams for each entry (i, j):
where x is a mapping x : [1, . . . , n] → [1, . . . , m]. That is, a vertical M(i, j) = e(i, j)+
seam is an 8-connected path of pixels in the image from top to bot- min(M(i − 1, j − 1), M(i − 1, j), M(i − 1, j + 1))
tom, containing one, and only one, pixel in each row of the image
(see Figure 1). Similarly, if y is a mapping y : [1, . . . , m] → [1, . . . , n],
At the end of this process, the minimum value of the last row in
then a horizontal seam is:
M will indicate the end of the minimal connected vertical seam.
Hence, in the second step we backtrack from this minimum entry on
y
sy = {s j }m = {( j, y( j))}m , s.t. ∀ j|y( j) − y( j − 1)| ≤ 1. (3)
j=1 j=1 M to find the path of the optimal seam (see Figure 1). The definition
of M for horizontal seams is similar.
The pixels of the path of seam s (e.g. vertical seam {si }) will there-
fore be Is = {I(si )}n = {I(x(i), i)}n . Note that similar to the
i=1 i=1
3.1 Energy Preservation Measure
removal of a row or column from an image, removing the pixels of
a seam from an image has only a local effect: all the pixels of the To evaluate the effectiveness of the different strategies for content-
image are shifted left (or up) to compensate for the missing path. aware resizing, we can examine the average energy of all of pixels
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
4. 10-4 • Avidan et al.
(b) e1
(a) Original
(c) eEntropy
Figure 5: Comparing aspect ratio change. From left to right in
the bottom: the image resized using seam removals, scaling and
cropping.
where HoG(I(x, y)) is taken to be a histogram of oriented gradients
at every pixel [Dalal and Triggs 2005]. We use an 8-bin histogram
computed over a 11 × 11 window around the pixel. Thus, taking
(d) eHoG (e) Segmentation and L1 the maximum of the HoG at the denominator attracts the seams to
edges in the image, while the numerator makes sure that the seam
Figure 4: Comparing different energy functions for content aware will run parallel to the edge and will not cross it. eHoG was also
resizing. used in Figures 9 and 10.
1
in an image |I| ∑ p∈I e(p) during resizing. Randomly removing pix- As expected, no single energy function performs well across all
images but in general they all accommodate a similar range for re-
els should keep the average unchanged, but content-aware resizing
sizing. They vary in the rate at which they introduce visual artifacts
should raise the average as it removes low energy pixels and keeps
and the parts of the image they affect. We found either e1 or eHoG
the high energy ones. Figure 3 shows a plot of the change in aver-
to work quite well.
age energy while changing the image width of Figure 2 using the
five different strategies outlined above. As expected, removing the
low energy pixels in ascending order gives the optimal result. This 4 Discrete Image Resizing
is closely followed by pixel removal. But both methods destroy the
visual coherence of the image. Cropping shows the worst energy 4.1 Aspect Ratio Change
preservation. Column removal does a better job at preserving en-
ergy, but still introduce visual artifacts. Seam carving strikes the Assume we want to change the aspect ratio of a given image I from
best balance between the demands for energy preservation and vi- n × m to n × m where m − m = c. This can be achieved simply by
sual coherency. This graph results are characteristic to many images successively removing c vertical seams from I. Contrary to simple
in general. scaling, this operation will not alter important parts of the image (as
defined by the energy function), and in effect creates a non-uniform,
3.2 Image Energy Functions content-aware resizing of the image (Figure 5).
We have examined several possible image importance measures The same aspect ratio correction, from n × m to n × m can also
found in literature as the energy function to guide seam carving. be achieved by increasing the number of rows by a factor of m/m
We have tested both L1 and L2 -norm of the gradient, saliency (Figure 6). The added value of such an approach is that it does not
measure [Itti et al. 1999], and Harris-corners measure [Harris and remove any information from the image. We discuss our strategy
Stephens 1988]. We also used eye gaze measurement [DeCarlo and for increasing an image size in details in sub-section 4.3.
Santella 2002], and the output of face detectors.
4.2 Retargeting with Optimal Seams-Order
Figure 4 compares the results of the e1 error, entropy, segmentation,
and Histogram of Gradients (HoG). The entropy energy computes Image retargeting generalizes aspect ratio change from one dimen-
the entropy over a 9×9 window and adds it to e1 . The segmentation sion to two dimensions such that an image I of size n × m will be
method first segments the image [Christoudias et al. 2002] and then retargeted to size n × m and, for the time being, we assume that
applies the e1 error norm on the results, effectively leaving only the m < m and n < n. This begs the question of what is the correct
edges between segments. Finally, eHoG is defined as follows: order of seam carving? Remove vertical seams first? Horizontal
seams first? Or alternate between the two? We define the search
∂ ∂
| ∂ x I| + | ∂ y I| for the optimal order as an optimization of the following objective
eHoG (I) = , function:
max (HoG(I(x, y))
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
5. Seam Carving for Content-Aware Image Resizing • 10-5
Figure 7: Optimal order retargeting: On the top is the original im-
Figure 6: Aspect ratio change of pictures of the Japanese master age and its transport map T. Given a target size, we follow the
Utagawa Hiroshige. In both examples the original image is widened optimal path (white path on T) to obtain the retargeted image (top
by seam insertion. row, right). For comparison we show retargeting results by alter-
nating between horizontal and vertical seam removal (top row, left),
removing vertical seams first (bottom row, left), and removing hor-
izontal seams first (bottom row, right)
k
min
sx ,sy ,α
∑ E(αi sx + (1 − αi )sy )
i i (5)
i=1
created after t seam have been removed from I. To enlarge an image
where k = r + c, r = (m − m ), c = (n − n ) and αi is used as a pa- we approximate an ‘inversion’ of this time evolution and insert new
rameter that determine if at step i we remove a horizontal or vertical ‘artificial’ seams to the image. Hence, to enlarge the size of an
seam: αi ∈ {0, 1} , ∑k αi = r , ∑k (1 − αi ) = c
i=1 i=1 image I by one we compute the optimal vertical (horizontal) seam
We find the optimal order using a transport map T that specifies, s on I and duplicate the pixels of s by averaging them with their left
for each desired target image size n × m , the cost of the optimal and right neighbors (top and bottom in the horizontal case).
sequence of horizontal and vertical seam removal operations. That Using the time evolution notation, we denote the resulting image as
is, entry T (r, c) holds the minimal cost needed to obtain an image
I(−1) . Unfortunately, repeating this process will most likely create
of size n − r × m − c. We compute T using dynamic programming.
a stretching artifact by choosing the same seam (Figure 8(b)). To
Starting at T(0, 0) = 0 we fill each entry (r, c) choosing the best of
achieve effective enlarging, it is important to balance between the
two options - either removing a horizontal seam from an image of
original image content and the artificially inserted parts. Therefore,
size n − r × m − c + 1 or removing a vertical seam from an image of
to enlarge an image by k, we find the first k seams for removal,
size n − r + 1 × m − c:
and duplicate them in order to arrive at I(−k) (Figure 8(c)). This
T(r, c) = min(T(r − 1, c) + E(sx (In−r−1×m−c )), can be viewed as the process of traversing back in time to recover
(6)
T(r, c − 1) + E(sy (In−r×m−c−1 ))) pixels from a larger image that would have been removed by seam
removals (although it is not guaranteed to be the case).
where In−r×m−c denotes an image of size n − r × m − c, E(sx (I))
and E(sy (I)) are the cost of the respective seam removal operation. Duplicating all the seams in an image is equivalent to standard
scaling (see Figure 8 (e)). To continue in content-aware fashion
We store a simple n × m 1-bit map which indicates which of the for excessive image enlarging (for instance, greater than 50%), we
two options was chosen in each step of the dynamic programming. break the process into several steps. Each step does not enlarge the
Choosing a left neighbor corresponds to a vertical seam removal size of the image in more than a fraction of its size from the pre-
while choosing the top neighbor corresponds to a horizontal seam vious step, essentially guarding the important content from being
removal. Given a target size n ×m where n = n−r and m = m−c, stretched. Nevertheless, extreme enlarging of an image would most
we backtrack from T(r, c) to T(0, 0) and apply the corresponding probably produce noticeable artifacts (Figure 8 (f)).
removal operations. Figure 7 shows an example of different retar-
geting strategies on an image. 4.4 Content Amplification
4.3 Image Enlarging Instead of enlarging the size of the image, seam carving can be used
to amplify the content of the image while preserving its size. This
The process of removing vertical and horizontal seams can be seen can be achieved by combining seam carving and scaling. To pre-
as a time-evolution process. We denote I(t) as the smaller image serve the image content as much as possible, we first use standard
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
6. 10-6 • Avidan et al.
(a) (b) (c) (d)
(e) (f) (g)
Figure 8: Seam insertion: finding and inserting the optimum seam on an enlarged image will most likely insert the same seam again and again
as in (b). Inserting the seams in order of removal (c) achieves the desired 50% enlargement (d). Using two steps of seam insertions of 50% in
(f) achieves better results than scaling (e). In (g), a close view of the seams inserted to expand figure 6 is shown.
Figure 9: Content amplification. On the right: a combination of
seam carving and scaling amplifies the content of the original image
(left).
scaling to enlarge the image and only then apply seam carving on
the larger image to carve the image back to its original size (see
Figure 9). Note that the pixels removed are in effect sub-pixels of Figure 10: Seam Carving in the gradient domain. The original
the original image. image (top left) is retargeted using standard technique (top right)
and in the gradient domain (bottom right). Zoom in comparison is
4.5 Seam Carving in the gradient domain shown on bottom left.
There are times when removing multiple seams from an image still
creates noticeable visual artifacts in the resized image. To over-
come this we can combine seam carving with Poisson reconstruc-
tion ([Perez et al. 2003]). Specifically, we compute the energy func-
tion image as before, but instead of removing the seams from the
original image we work in the gradient domain and remove the
seams from the x and y derivatives of the original image. At the
end of this process we use a poisson solver to reconstruct back the
image. Figure 10 shows an example of this technique.
4.6 Object Removal
We use a simple user interface for object removal. The user marks
the target object to be removed and then seams are removed from
the image until all marked pixels are gone. The system can auto-
matically calculate the smaller of the vertical or horizontal diame-
ters (in pixels) of the target removal region and perform vertical or
horizontal removals accordingly (Figure 11). Moreover, to regain Figure 11: Simple object removal: the user marks a region for re-
the original size of the image, seam insertion could be employed on moval (green), and possibly a region to protect (red), on the original
the resulting (smaller) image (see Figure 12). Note that, contrary image (see inset in left image). On the right image, consecutive ver-
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007. seam were removed until no ‘green’ pixels were left.
tical
7. Seam Carving for Content-Aware Image Resizing • 10-7
Figure 14: Retargeting the left image with e1 alone (middle), and
with a face detector (right).
Figure 12: Object removal: find the missing shoe! (original image
is top left). In this example, in addition to removing the object (one
shoe), the image was enlarged back to its original size. Note that
this example would be difficult to accomplish using in-painting or
texture synthesis.
Figure 13: An image with its vertical and horizontal seam index
maps V and H, colored by their index from blue (first seams) to red
(last seams).
Figure 15: Retargeting the Buddha. At the top is the original image,
to previous object removal techniques [Drori et al. 2003; Criminisi a cropped version where the ornaments are gone, and a scaled ver-
et al. 2003; Bertalmio et al. 2003], this scheme alters the whole im- sion where the content is elongated. Using simple bottom up fea-
age (either its size or its content if it is resized). This is because both ture detection for automatic retargeting cannot protect the structure
the removed and inserted seams may pass anywhere in the image. of the face of the Buddha (Bottom, left) and this is a challenging
image for face detectors as well. By adding simple user constraints
to protect the face (Bottom, middle) or the face and flower (Bottom,
5 Multi-size Images right), better results are achieved.
So far we have assumed that the user knows the target size ahead
of time, but this might not be possible in some cases. Consider, moval (Figure 13). To get an image of width m , we only need to
for example, an image embedded in a web page. The web designer gather, in each row, all pixels with seam index greater than or equal
does not know, ahead of time, at what resolution the page will be to m − m .
displayed and therefore, cannot generate a single target image. In a This representation supports image enlarging as well as reduction.
different scenario, the user might want to try different target sizes For instance, if we want to support enlarging of the image up to size
and choose the one most suitable for his or her needs. M > m, we enlarge the image using seam insertion procedure to a
Seam carving is linear in the number of pixels and resizing is there- size n × M similar to Section 4.3. However, instead of averaging
fore linear in the number of seams to be removed, or inserted. On the k-th seam with it’s two neighbors, we do not modify the original
average, we retarget an image of size 400 × 500 to 100 × 100 in image pixels in the seam, but insert new pixels to the image as the
about 2.2 seconds. However, computing tens or hundreds of seams average of the k-th seam and its left (or right) pixel neighbors. The
in real time is a challenging task. To address this issue we present inserted seams are given a negative index starting at −1. Conse-
a representation of multi-size images that encodes, for an image of quently, to enlarge the original image by k, (m < k ≤ M ), we use
size (m × n), an entire range of retargeting sizes from 1 × 1 to m × n exactly the same procedure of gathering (from the enlarged image)
and even further to N × M , when N > n, M > m. This informa- all pixels whose seam index is greater than (m − (m + k)) = −k, and
tion has a very low memory footprint, can be computed in a couple get an image of size m − (−k) = m + k.
of seconds as a pre-processing step, and allows the user to retaget
an image continuously in real time. Computing a horizontal index map H for image height enlarging
and reduction is achieved in a similar manner (see Figure 13). How-
From a different perspective, this can be seen as storing an explicit ever, supporting both dimension resizing while computing H and V
representation of the time-evolution implicit process of seam re- independently will not work. This is because horizontal and verti-
movals and insertions. Consider, first, the case of changing the cal seams can collide in more than one place, and removing a seam
width of the image. We define an index map V of size n × m that in one direction may destroy the index map in the other direction.
encodes, for each pixel, the index of the seam that removed it, i.e., More details can be found in the appendix. However, a simple way
V(i, j) = t means that pixel (i, j) was removed by the t-th seam re- to avoid this is to allow seam removal in one direction, and use de-
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
8. 10-8 • Avidan et al.
Acknowledgments
We would like to thank Fr´ do Durand and the graphics
e
group at MIT for reviewing an early version of this work.
We thank Stark Draper for narrating the video. We thank
Eric Chan for the use of the waterfall image, and numer-
ous flickr (http://www.flickr.com/) members for mak-
ing their images available through creative common rights
(http://creativecommons.org/): crazyegg95 (Buddha),
Gustty (Couple and Surfers), JeffKubina (Capitol), mykaul (Han-
nuka and Car), o2ma (Vase), sigs66 (Long beach and Two people
near sea). We also thank the anonymous reviewers and referees for
their comments.
Figure 16: Examples when resizing using seams fails: images that
are too condensed (left) or where the content layout prevents seams References
to bypass important parts (right). In such cases the best strategy
would be to use scaling.
AGARWALA , A., D ONTCHEVA , M., AGRAWALA , M., D RUCKER ,
S., C OLBURN , A., C URLESS , B., S ALESIN , D., AND C OHEN ,
generate seams, i.e. rows or columns, in the other direction. Note M. 2004. Interactive digital photomontage. ACM Trans. Graph.
that although retargeting a multi-size image to any size is instanta- 23, 3, 294–302.
neous, due to the additional constraints, the resulting image would
be different than the one created in the optimal order induced by B ERTALMIO , M., S APIRO , G., C ASELLES , V., AND BALLESTER ,
the implicit time-evolving process of subsection 4.2. The reader is C. 2000. Image inpainting. In Proceedings of ACM SIGGRAPH,
referred to the attached video for example of continuously resizing 417–424.
multi-size images to various sizes in real-time. B ERTALMIO , M., V ESE , L., S APIRO , G., AND O SHER , S. 2003.
Simultaneous structure and texture image inpainting. In Proc.
6 Limitations IEEE Conference on Computer Vision and Pattern Recognition,
707–714.
All the examples shown so far in this paper were computed au- B OYKOV, Y., AND J OLLY, M.-P. 2001. Interactive graph cuts for
tomatically, but our method does not work automatically on all im- optimal boundary & region segmentation of objects in n-d im-
ages. This can be corrected by adding higher level cues, either man- ages. In International Conference on Computer Vision, (ICCV),
ual or automatic. For example, in Figure 14 the e1 error function vol. I, 105–112.
fails, but combined with a face detector we get much better results.
Figure 15 shows an example of adding user constraints. C HEN , L., X IE , X., FAN , X., M A , W., Z HANG , H., AND Z HOU ,
H. 2003. A visual attention model for adapting images on small
Other times, not even high level information can solve the problem. displays. Multimedia Systems 9, 4, 353–364.
We can characterize two major factors that limit our seam carving
approach. The first is the amount of content in an image. If the C HRISTOUDIAS , C., G EORGESCU , B., AND M EER , P. 2002. Syn-
image is too condensed, in the sense that it does not contain ‘less ergism in low-level vision. In 16th International Conference on
important’ areas, then any type of content-aware resizing strategy Pattern Recognition, vol. IV, 150–155.
will not succeed. The second type of limitation is the layout of the C RIMINISI , A., P EREZ , P., AND T OYAMA , K. 2003. Object re-
image content. In certain types of images, albeit not being con- moval by exemplar-based inpainting. In In IEEE Conference on
densed, the content is laid out in a manner that prevents the seams Computer Vision and Pattern Recognition, 417–424.
from bypassing important parts (Figure 16).
DALAL , N., AND T RIGGS , B. 2005. Histograms of oriented gradi-
ents for human detection. In International Conference on Com-
7 Conclusions and Future Work puter Vision & Pattern Recognition, vol. 2, 886–893.
We presented an operator for content-aware resizing of images us- DAVIS , J. 1998. Mosaics of scenes with moving objects. In Pro-
ing seam carving. Seams are computed as the optimal paths on ceedings of CVPR.
a single image and are either removed or inserted from an image.
This operator can be used for a variety of image manipulations in- D E C ARLO , D., AND S ANTELLA , A. 2002. Stylization and ab-
cluding: aspect ratio change, image retargeting, content amplifica- straction of photographs. In Proceedings of SIGGRAPH, 769–
tion and object removal. The operator can be easily integrated with 776.
various saliency measures, as well as user input, to guide the resiz- D RORI , I., C OHEN -O R , D., AND Y ESHURUN , Y. 2003. Fragment-
ing process. In addition, we define a data structure for multi-size based image completion. In Proceedings of ACM SIGGRAPH,
images that support continuous resizing ability in real time. 303–312.
There are numerous possible extensions to this work. We would E FROS , A. A., AND F REEMAN , W. T. 2001. Image quilting for
like to extend our approach to other domains, the first of which texture synthesis and transfer. In SIGGRAPH 2001, Computer
would be resizing of video. Since there are cases when scaling can Graphics Proceedings, ACM Press / ACM SIGGRAPH, E. Fi-
achieve better results for resizing, we would like to investigate the ume, Ed., 341–346.
possibility to combine the two approaches, specifically to define
more robust multi-size images. We would also like to find a better G AL , R., S ORKINE , O., AND C OHEN -O R , D. 2006. Feature-
way to combine horizontal and vertical seams in multi-size images. aware texturing. In Eurographics Symposium on Rendering.
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.
9. Seam Carving for Content-Aware Image Resizing • 10-9
H ARRIS , C., AND S TEPHENS , M. 1988. A combined corner and
edge detector. In Proceedings of the 4th Alvey Vision Conference,
147–151.
I TTI , L., KOCH , C., AND N EIBUR , E. 1999. A model of saliency-
based visual attention for rapid scene analysis. PAMI 20, 11,
1254–1259.
JACOBS , C., L I , W., S CHRIER , E., BARGERON , D., AND Figure 17: The horizontal and vertical consistent index maps of the
S ALESIN , D. 2003. Adaptive grid-based document layout. In image on figure 13. Color index goes from blue (first seams) to red
Proceedings of ACM SIGGRAPH, 838–847. (last seams).
J IA , J., S UN , J., TANG , C.-K., AND S HUM , H.-Y. 2006. Drag- A Constructing Consistent Index Maps
and-drop pasting. In Proceedings of SIGGRAPH.
K UHN , H. W. 1955. The hungarian method for the assignment Computing an horizontal index map H and vertical index map V in-
problem. In Naval Research Logistic Quarterly, 2:83–97. dependently for multi-size image will not work. We say that H and
V are consistent if every horizontal seam intersects (or touches) all
L IU , F., AND G LEICHER , M. 2005. Automatic Image Retargeting the vertical seam indexes and every vertical seam intersects all hor-
with Fisheye-View Warping. In ACM UIST, 153–162. izontal seam indexes. Consistency assures that removing a seam in
any dimension will remove exactly one pixel from all seams in the
L IU , F., AND G LEICHER , M. 2006. Video Retargeting: Automat- other dimension, retaining the index map structure. If consistency is
ing Pan and Scan. In ACM international conference on Multime- not maintained, then after removing one horizontal seam we might
dia, 241–250. be left with vertical seams with different number of pixels and the
rectangular structure of the image will be destroyed.
L IU , H., X IE , X., M A , W., AND Z HANG , H. 2003. Automatic
browsing of large pictures on mobile devices. Proceedings of Aside from limiting seams to be rows or columns in one, or both
the eleventh ACM international conference on Multimedia, 148– dimensions, we present here another approach. We use only tem-
155. porally 0-connected seams, i.e. seams that are spatially connected
on the original size image I(0) . For such seams, the only possi-
P EREZ , P., G ANGNET, M., AND B LAKE , A. 2003. Poisson image ble violation of consistency between the H and V maps can occur
editing. ACM Trans. Graph. 22, 3, 313–318. in diagonal seam steps. Our method first computes temporally 0-
connected seams in one direction, and then imposes the constraints
ROTHER , C., B ORDEAUX , L., H AMADI , Y., AND B LAKE , A. on the diagonal when computing the seams in the other direction.
2006. Autocollage. In Proceedings of SIGGRAPH 2006.
To understand why the only violation of consistency occurs in di-
S ANTELLA , A., AGRAWALA , M., D E C ARLO , D., S ALESIN , D., agonals, assume without loss of generality, that some vertical seam
AND C OHEN , M. 2006. Gaze-based interaction for semi- j ∈ {1 . . . m} violates the consistency constraint. This means it must
automatic photo cropping. In ACM Human Factors in Comput- touch some horizontal seam i ∈ {1 . . . n} more than once. Denote
ing Systems (CHI), 771–780. those pixels where seam j meets seam i as p and q. Since p and q
are part of a vertical seam j, they cannot be in the same row. How-
S ETLUR , V., TAKAGI , S., R ASKAR , R., G LEICHER , M., AND ever, they are also part of the horizontal seam i, and cannot be in the
G OOCH , B. 2005. Automatic Image Retargeting. In In the same column. Let us examine the rectangle defined by p and q in its
Mobile and Ubiquitous Multimedia (MUM), ACM Press. corners. seams i and j must be connected inside this rectangle and
S UH , B., L ING , H., B EDERSON , B. B., AND JACOBS , D. W. they both touch it’s corners. However, one is a vertical seam and
2003. Automatic thumbnail cropping and its effectiveness. In the other a horizontal seam. The only possibility for this to happen
UIST ’03: Proceedings of the 16th annual ACM symposium on is that the rectangle is in fact a square, and both seams pass through
User interface software and technology, ACM Press, New York, its diagonal.
NY, USA, 95–104. Note that the above claim relies on the fact that all seams are con-
nected in the original image, which is not true if we use non 0-
S UN , J., Y UAN , L., J IA , J., AND S HUM , H. 2005. Image com- connected seams. However, because we are using 0-connected
pletion with structure propagation. In Proceedings of ACM SIG- seams, we can simultaneously compute all of them in one direc-
GRAPH. tion. Without loss of generality, for 0-connected vertical seams we
examine all pairs of rows of the original image independently. For
V. K WATRA , A. S CHDL , I. E. G. T., AND B OBICK , A. 2003.
each such pair we find the optimal set of 1-edge paths linking all
Graphcut textures: Image and video synthesis using graph cuts.
pixels of one row to all pixels of the next row. The global multiple
In Proceedings of SIGGRAPH.
seam paths from the top of the image to its bottom would simply be
V IOLA , P., AND J ONES , M. 2001. Rapid object detection using a the concatenation of those 1-edge paths.
boosted cascade of simple features. In Coference on Computer Finding the best 1-edge paths between a pair of rows is similar to a
Vision and Pattern Recognition (CVPR). weighted assignment problem where each pixel in one row is con-
nected to its three neighboring pixels in the other row. We use the
WANG , J., AND C OHEN , M. 2006. Simultaneous Matting and
Hungarian algorithm [Kuhn 1955] to solve this weighted assign-
Compositing. Microsoft Research Technical Report, MSR-TR-
ment problem. Once we find the seams in one direction, we repeat
2006-63 (May).
the process in the other direction, but we mask out every diagonal
Z OMET, A., L EVIN , A., P ELEG , S., AND W EISS , Y. 2005. Seam- edge that was already used by any of the first direction seams. This
less image stitching by minimizing false edges. IEEE Transac- guarantees that the seams in the second direction will be consistent
tions on Image Processing 15, 4, 969–977. with the first direction (Figure 17).
ACM Transactions on Graphics, Vol. 26, No. 3, Article 10, Publication date: July 2007.