Published on

Published in: Technology, News & Politics
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide


  1. 1. International Journal of ElectronicsJOURNAL OF ELECTRONICS AND INTERNATIONAL and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET) ISSN 0976 – 6464(Print) ISSN 0976 – 6472(Online) Volume 5, Issue 1, January (2014), pp. 52-58 © IAEME: www.iaeme.com/ijecet.asp Journal Impact Factor (2013): 5.8896 (Calculated by GISI) www.jifactor.com IJECET ©IAEME IMAGE RETRIEVAL AND FACE RECOGNITION TECHNIQUES: LITERATURE SURVEY Mrs. Manisha Bhisekar M.E. - 2nd year -E&TC -Digital Systems, G.S. Moze C.O.E., Balewadi, Pune Prof. Prajakta Deshmane ABSTRACT Now days the process of image retrieval is widely used in may real life applications from large datasets. This process of retrieving the images from the big images dataset is called as content based image retrieval (CBIR). These CBIR techniques are adopted recently in many image based applications like fingerprint matching image retrieval, face based image retrieval, and attribute based image retrieval. The area of interest during this review paper is face based image retrieval. We have studied many methods presented recently for the face based image retrieval in which different face recognition algorithms were used. The goal of face based image retrieval is to display the face results those are exactly related to the query image of person. In this paper, first we present the review of CBIR methods, after that different face recognition methods discussed with their advantages and disadvantages. Keywords: Image Retrieval, CBIR, Face-Based Image Retrieval, Face Recognition, Query Image. I. INTRODUCTION Since from last decade, there was tremendous growth in use of technology and science in different fields which is resulted into the huge amount of images in different areas such as art galleries, nature, entertainment, education, industry, biomedical, security etc. In these applications, we frequently need to store and retrieve for various purposes, especially for decision making. Retrieving the particular image from huge image dataset is very complex and tedious tasks; therefore there must be an automated system which can able to retrieve number matching images to the input query image. The designing of such systems is quite challenging in research. There are many image retrieval systems presented so far for automatic retrieval of images from large image dataset. There are two types of image retrieval systems such as content based and text based image retrieval. For the 52
  2. 2. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME text based system, text annotation is done manually for all images and then used by a database management system to perform image retrieval. This manual process may takes more time for doing so. There are two main limitations of this approach such as more resources and costs are required to do the manual image annotations and the process of explaining the contents of image highly subjective [1]. That is, the perspective of textual descriptions given by an annotator could be different from the perspective of a user. In other words, there are textual user queries and image annotation or description is inconsistencies between. Incompatibility to correct the problem, the image according to image content retrieval is carried out such a strategy the so-called content-based image Retrieval (CBIR). Efficient and effective retrieval [1], [2] to facilitate the images to build meaningful description of the physical characteristics of CBIR system is the primary goal of CBIR system basically features Extraction and matching these features works on each dataset with queries image in color, shape and texture features. Based on this extracted features contents are matched and hence the results generated automatically without any manual work. Later this approach of CBIR is used in under different real time environments such as security, medical imaging, photography etc. Recently this concept is well utilized in face based image retrieval method. The face based image retrieval methods are nothing but the combination of face recognition and CBIR techniques. This method is basically used for the extraction of persons images from large dataset based on input query image of same person. Internally this process works same As CBIR, only its uses the extra features like face detection, face recognition and face alignment techniques. A first, though very limited, approach to achieve this goal has been implemented in the Photo Book system [8] which allowed for retrieving images of persons when the face covered a large part of the image and was taken under normalized conditions. In [2] an approach to the same problem is proposed. Portrait images, glasses, hats, facial expressions and facial hair with strong respect for retrieving a more complex approach [7] is presented in a component-based descriptors for images using LDA face changes [11] is presented in [1] portrait images taken from the Web are indexed in such systems are proposed for comparing performance measures. These methods have in common is that all of the image same face. In addition it assumes that to be recognized is the image of the face, the General image retrieval is not an appropriate assumption for a significant portion forms. In contrast to these approaches we propose a method that is able to deal with images displaying several persons and in which the faces do not necessarily form large parts of the image. In this paper we are presenting our review of two different approaches, below in section II we will discuss about CBIR system and its different aspects, in section III different face recognition, detection methods used for face based image retrieval systems. This paper is our roadmap for future work. II. REVIEW OF IMAGE RETRIEVAL TECHNIQUES With the rapid growth of digital devices for capturing and storing multimedia data, Multimedia information retrieval is one of the most important research subjects among which image retrieval key challenge has been one of the problems in recent years. in content-based image retrieval (CBIR) which image retrieval in the last decade many research interests in computer communities has attracted a wide range of the most important topics [6]. Although extensive studies have been conducted, finding desired images from multimedia databases is still a challenging and open issue. The main challenges are due to two gaps in CBIR [6]. The first is the sensor gap between the object of the world and the information represented by computers. The second one is the semantic gap between the low-level visual features and high-level human perception and interpretation. Many early year studies on CBIR focused primarily on feature analysis which mainly aimed at solving the sensory gap [8], [7]. 53
  3. 3. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME However, because of the complexity of image understanding and the challenge of semantic gap, it is impossible to discriminate all images by employing some rigid simple similarity measure on the low-level features. Although it is feasible to bridge the semantic gap by building an image index with textual descriptions, manual indexing on image databases is typically time consuming, costly and subjective, and hence difficult to be fully deployed in practical applications. Despite the promising process recently reported in image annotations [9], fully automatic image annotation is still a long way off. Relevance feedback, as an alternative and more feasible technique to mitigate the semantic gap issue [10]. Increasing developments in computer and communication technologies have led to enormous archives of digital images in various areas such as digital library, medicine, art galleries, remote sensing, education, entertainment and so on. That is why transfer of digital images has been considered as one of the most important research topics for more than 40 years and today most of the efforts are that firstly to decrease the volume of storage and secondly to retrieve the proposed images with acceptable speed and precision rate. In this area a wide range of researches revolves on search based techniques on image databases as a critical need. The primary search-based solutions depend on textural descriptions provided by human operators which two problems were associated with this approach: expensiveness and low efficiency. The first one comes out of the amount of manual labor time required for image annotation and the second one comes out of the fact that each image may contain several. Before moving to in details about the CBIR systems, below we will discuss about the image retrieval system: 2.1 Image Retrieval: Techniques of image retrieval integrate both addressing the more detailed perceptual aspects, low-level visual features and high-level semantic features which the more detailed concept of the visual data Public and the private entities controls the rapid growth of the number and the types of the assets of the multimedia which are get controlled by the entities which are public and private, and also they are expanding the range of the video and the image documents which are appearing on the web, they provide very good tools for retrieving very perfect and good image visual data. The process of image retrieval is perfectly depends on the availability of the image contents. The descriptor of the image is the visual features like texture, shape, color and somatic primitives and spatial relationships. Image feature retrieval of text image based solely on recovery values are conveniently represented as a vector. However, "a picture is worth a thousand words." images, text content, content and more versatile view of data is very large and still expanding very rapidly presenting the Visual data. to cope with the new system of content-based image retrieval system The family of image-based content is named recovery system high level low level of semantic features, view features and more detailed addressing the conceptual aspects. For retrieving and managing the visual data sufficient for two types of features only but it is not only case but researches face one problem that not only to retrieve the features but combine those features also this is huge barrier in the front of the researchers. The satisfactory performance is not provided by the heuristic approaches and intuitive approach. Therefore, mange low level features and high level concepts we need to managing latent correlation. The main challenge in front of the researchers is that how to manage this bridge between the somatic features and visual features. The images associates’ different type of information with them this information is: The different types of information that are normally associated with images are: • Metadata Content-independent: data is related to the image content but is not concerted with image Examples are author’s name, image format, date, and location. 54
  4. 4. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME • - - Metadata Content-based: Metadata Non-information-bearing: the data that referring to the intermediate level or to the low-level features such as texture color. Shape, relationships, data referring to low-level or intermediate-level features, such as color, texture, shape, spatial relationships, and various combinations for them. Such information is computed by using the row data. Metadata Information-bearing: semantic are referred by the data, real work entities and concerned with relationships. For this type of the method the is Empire State Building it is such as particular building which are appear into the image, this building is cannot derived from raw data, must be supplied by the other means, perhaps inheriting the this semantic label from the another image, where a similar appearing building has already been identified. 2.2 Existing Techniques: Extraction visual features are the basis of any content-based image retrieval technique. Basically used features such as include texture, color, shape and special relationship. Due to composition of complex visual data and subjectivity of perception ,there is actually not exist single best representation for given feature visual, Various approaches is introduced for each of these visual features and each of them characterizes the feature from a different perspective. In content-based image retrieval, color is one of the mostly used visual features. it is very easy and robust to represent. Various studies of spaces color and perception color is proposed, in order to find out color based techniques that are mostly closed to human color perceive align. The color histogram most widely used presentation technique, statistically describing from combined probabilistic properties of the various color channels by capturing the number of pixels having particular properties. For example, a color histogram might describe the number of pixels of each red channel value in the range [0, 255]. Three of its derived color histograms, particular channels values is shown along such as x-axis and numbers of pixels shown along with yaxis and particular color channels used that indicated in each histogramIt's a well-known such as histograms information related to spatial distribution of colors and two very different images can be very similar histograms. There is more work to capture such local histograms information has been extending more spatial detail information, Histograms. Correlograms and angiograms are the two main approaches used to capture such as Correlograms. Special areas around the pixels in special colors pixel Distribution of colors is used to capture and angiograms is used for General properties of spatial arrangement of a special signature capture. Angiograms texture and size can also be used for features. 2.3 Content-Based Image Retrieval (CBIR) Systems: There are various excellent surveys with CBIR (Content based image). QBIC (Query-by-Image-Content) was one of the first prototype systems. QBIC was developed at the IBM Alma den Research Center and now is currently folded into DB2. It is used to allow queries by shape, texture, and color, and introduce a sophisticated similarity function. Similarity function has a quadratic time-complexity. To speed-up searches, multidimensional indexing is used is another property of QBIC. The MARS system is developed at the University of Illinois at Urbana-Champaign, and that allows for sophisticated relevance feedback from the user. III. REVIEW OF FACE RECOGNITION METHODS To allow for investigating the effect of face detection and representation methods, we tested two methods: detection and representation using Eigenfaces [10] and detection using the Viola & Jones method [11] and representing the face as a size normalized image patch. Both methods are data driven methods and are explained briefly in the following. A general overview on methods for face detection and representation can be found in [12]. 55
  5. 5. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME 3.1 Eigenfaces Turk and Pent land applied principal component analysis to face recognition and detection [10]. Principal component analysis is performed on a training set of face images to generate the Eigen-vectors (here called Eigenfaces) which span a subspace (called the face space) of the image space. To determine whether an image is a face or not, it is estimated, for instance subspace and back only 20 first face in using the space components. Again, the original image and back-projection can calculate the distance between. Due to the nature of the Eigenfaces, like face an image, while gaiSufaces are very poor and thus reconstructed the original image and back-projection is high in this case, the distance between the reconstructed. Thus, it is a measure of the distance as faceness. If faceness is calculated for every position in the image, a face can occur and a face map to find local minima can be traced from an advantage of this method is that it is a compact and generalizing the method of representing faces offers. 3.2 Viola & Jones Method Viola and Jones present a new and radically faster approach to face detection based on the AdaBoost algorithm from machine learning [11]. Boosting is a method of combining several weak classifiers to generate a strong classifier. Weak classifiers AdaBoost algorithm on statistical training and generalization error bounds is providing strong classifiers to produce a well-known algorithm. weak classifiers characteristics of three types in Viola & Jones algorithm are based on a tworectangle SuSum of the values of two adjacent rectangular Windows mode is the difference between a three-rectangle feature in three adjacent rectangles understands and extreme rectangles pixels in the center of Yoga and yogic rectangle computes the difference between. A four-set of 2 × 2 rectangle feature rectangles considers and calculates that main diagonals forming rectangles and the difference between the amounts of pixels off a 24 × 14 sub-window there are more than 180,000 features. 3.3 Faces in Image Retrieval To be able to use the faces in our image retrieval framework, we have to define a distance measure for two images X and Y in which faces X1 . . . XF and Y1 . . . YV have been detected. One can think about various ways of matching, e.g. taking into account face positions. For experiments presented here, we match the most simple to use as possible. That’s it, we're all pairs x and two images (Xi, Yj) between faces Y Euclidean distance d (Xi, Yj) calculate the distance d (Xi, Yj). The small distance between X and Y of the images used in the images of individuals in a query image Retrieving that can be interpreted as, and this person's face as the same query as possible to face it in the image. IV. DATABASE USED We are showing experiments multiple databases: a) Bio-ID database: a database present one person for every image which can say to be recorded in controlled conditions. b) RWTH-i6 Groups of People Database: a newly created database which has been collected using Google image search. It is obvious that this is a much harder task. 4.1 Bio-ID The dataset consists of 1521 gray level images with a resolution of 384×286 pixel. Each image shows the frontal view of a face of one out of 23 different test persons. The database is available online. This task is very simple and the results appear to be very good. Unfortunately, the data are not properly labeled, thus a quantitative evaluation is not easily possible. 56
  6. 6. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME 4.2 RWTH-i6 Groups of People Database Images of persons, single persons, and individuals without all of the images of groups of images due to the lack of a database containing, we decided to make our own database. To do this we politicians, musicians and music bands (such as Gerhard Schroeder, Britney Spears, Depeche Mode) using the names of Google Image Search FAQ and took each of the 60 images for these keywords. in total we used 38 search terms then We all have images that the search term, such as images showing dogs, comics, or other people would show up as images were deleted not relevant to it containing 867 images with 38 classes led to a database. V. CONCLUSION AND FUTURE WORK In this paper we have presented the review of image retrieval system and face based image retrieval system, especially the techniques of face detection and representation in image based face recognition system. We explained detailed working of content based image retrieval system as well as face based image retrieval system. In the literature there are many methods presented for face based image retrieval systems. But each method is suffered from different limitations. As per our study ignore strong, the lack of a specific geometric face-face image in view between words. Face recognition works in various recent discriminative facial features is proposed, but these features typically not global, high-dimensional and thus is suitable for quantization and inverted indexing. Future work we present efficient new method said on the problems and suggestions for the above to work. V. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] P. Anick, “Using Terminological Feedback for Web Search Refinement: A Log-Based Study,” Proc. 26th Ann. Int l ACM SIGIR Conf., pp. 88-95, 2003. Kekre, H.B. and Sudeep D. Thepade, 2008. Creating the Color Panoramic View using Medley of Grayscale and Color Partial Images, WASET International Journal of Electrical Computer and System Engineering (IJECSE), 2: 3. Kekre, H.B. and Sudeep D. Thepade, 2008. Color Traits Transfer to Gray scale Images, In Proc of IEEE First International Conference on Emerging Trends in Engineering & Technology. Kekre, H.B. and Sudeep D. Thepade, 2008. Scaling Invariant Fusion of Image Pieces in Panorama Making and Novel Image Blending Technique, International Journal on Imaging (IJI), 1(A08): 31-46. Müller, H., N. Michoux, D. Bandon and A. Geissbuhler, 2004. A review of content-based image Retrieval systems in medical applications – clinical benefits and future directions, International Journal Med. Inf., 73(1): 1-23. Lehmann, T., B. Wein, J. Dahmen, J. Bredno, F. Vogelsang and M. Kohnen, 2000. ContentBased Image Retrieval in Medical Applications: A Novel Multi-Step Approach, International Society for Optical Engineering, 3972(32): 312-320. Rapha¨el Mar´ee, Pierre Geurts, Justus Piater, and Louis Wehenkel. Random subwindows for robust image classification. In Cordelia Schmid, Stefano Soatto, and Carlo Tomasi, editors, IEEE Conference on Computer Vision and Pattern Recognition, volume 1, pages 34–40, San Diego, CA, USA, June 2005. IEEE. A. Pentland, R.W. Picard, and S. Sclaroff. Photobook: Content-based manipulation of image databases. International Journal of Computer Vision, 1995. David McG. Squire, Wolfgang M¨uller, Henning M¨uller, and Jilali Raki. Content-based query of image databases, inspirations from text retrieval: Inverted files, frequency-based 57
  7. 7. International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 5, Issue 1, January (2014), © IAEME [10] [11] [12] [13] [14] [15] [16] weights and relevance feedback. In Scandinavian Conference on Image Analysis, pages 143–149, Kangerlussuaq, Greenland, June 1999. Matthew Turk and Alex Pentland. Eigenfaces for recognition. Journal of Cognitive Neuroscience, 3(1):71–86, jan 1991. Paul Viola and Michael Jones. Robust real-time object detection. In Second International Workshop on Statistical and Computational Theories of Vision – Modeling, Learning, Computing, and Sampling, pages 1–25, Vancouver, Canada, July 2001. Ming-Hsuan Yang, David J. Kriegman, and Narendra Ahuja. Detecting faces in images: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence, 24(1):34–58, Jan 2002. Madhubala Myneni and Dr.M.Seetha, “Feature Integration for Image Information Retrieval using Image Mining Techniques”, International Journal of Computer Engineering & Technology (IJCET), Volume 3, Issue 3, 2012, pp. 273 - 281, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375. Tarun Dhar Diwan and Upasana Sinha, “Performance Analysis is Basis on Color Based Image Retrieval Technique”, International Journal of Computer Engineering & Technology (IJCET), Volume 4, Issue 1, 2013, pp. 131 - 140, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375. Abhishek Choubey, Omprakash Firke and Bahgwan Swaroop Sharma, “Rotation and Illumination Invariant Image Retrieval using Texture Features”, International Journal of Electronics and Communication Engineering & Technology (IJECET), Volume 3, Issue 2, 2012, pp. 48 - 55, ISSN Print: 0976- 6464, ISSN Online: 0976 –6472. K.Ganapathi Babu, A.Komali, V.Satish Kumar and A.S.K.Ratnam, “An Overview of Content Based Image Retrieval Software Systems”, International Journal of Computer Engineering & Technology (IJCET), Volume 3, Issue 2, 2012, pp. 424 - 432, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375. 58