Motion detection in compressed video using macroblock classification
Artifacts Detection by Extracting Edge Features and Error Block Analysis from Broadcasted Videos
1. Edge-Based Feature Extraction for Artifacts Detection and Error
Pattern Analysis from Broadcasted Videos
Supervised by
Prof. Oksam Chae
Md. Mehedi Hasan, 2010315443
Image Processing Lab,
Department of Computer Engineering
Kyung Hee University, Korea
2012.05.08
2. Presentation Outline
2
•Objectives
•Challenges
Introduction
Contributions
Related Works
•The Proposed Video Artifacts measure and Error Frame Detection
•The Proposed Spatial Error Block Analysis System
Proposed Artifact Detection and Error Pattern Analysis
Experimental Results
Conclusion and Future Work
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos
3. 3
Introduction
• To gain a system that detect video artifacts happened not only in compression
(Block based) but also occurred during transmission or broadcasting.
• To reduce the time complexity of the conventional pixel based detection
methods which requires high memory and too much computation time.
• Selection of light weighted human vision measurement system and Choosing a
detection mechanism to detect the distorted frames in real-time.
• Introduce a error block classification and analysis method that can be used in
video restoration, error concealment , video retrieval and many other
commercial applications .
Objective
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos
Noise and error model for broadcasting and surveillance systems
4. Introduction
4
Video artifacts detection and distorted pattern analysis is difficult
Videos are distorted with compression , wireless transmission based and broadcasting based artifacts
In image and video communication , original image and video is not accessible which is called No-reference approach, is a
challenging research issue.
Compression based artifacts are sustained in a block based manner (typically , 8 by 8) but wireless transmission and
broadcasting related artifacts are not always sustained in a block based manner.
A real time application that not only show the quality measure but also detect the distorted frames from videos.
Classify and analyze the error patterns from defected frames that can be used in video restoration, error concealment and
retrieval.
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos
Challenges
5. Introduction
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 5
Sample Videos and Images
Courtesy: Samples provided by KBS
6. Proposed Method
6
• Contains Edge magnitude and direction
• Less Sensitive to illumination Variation and Noise
• Extract frames from videos and analyze
• Incorporate Kirsch Mask to detect edge pixels.
• Can detect candidate frame with high disruption in
sequence of frames(Temporal Information).
• More gradient direction is analyzed for complex
environment
• Block classification is done in three steps.
• Edge Block and Texture Content Block is analyzed .
• Error Block Analysis is incorporated for better accuracy
that can be used is Error concealment and restoration.
Video Artifacts Measure and Error
Frame Detection
Spatial Error Block Analysis (SEBA)
Statistical Background Modeling and Multiple Motion Analysis for the Parametric Gesture Representation
8. Artifact Measure and Error Frame Detection
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 8
9. Generate Distortion Metric
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 9
Further the locations of the
compression block boundaries may
be detected by observing where the
maximum correlation value occurs.
The resulting correlation results are
proposed to generate a picture
quality rating for the image which
represents the amount of human-
perceivable block degradation that
has been introduced into the
proposed video signal.
Combining the results in a simple
way yields a metric that shows a
promising performance with respect
to practical reliability, prediction
accuracy, and computational
efficiency.
10. Distortion Metric for Error Frames
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos10
11. Error Frame Detection
To compute the distortion measure of every frame we compare deviation with the
previous frame. If the value is within a certain threshold value then it is considered
as successful undistorted frame. Otherwise it is consider as distorted frame and
forwarded to next report results module.
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos11
0
N
r msr
n
F B n
Criteria
Function
Deviation of
Frames
Calculate
Mean of
frames
12. Spatial Error Block Analysis
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos12
Proposed System
13. Flowchart of Block Analysis
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos13
Edge Direction
Classification
Error Block Classification
Block Shape and
Rotation Formulation
Forward parameters for
error concealment
Detected Error Frame
Sobel Mask in 60
Gradient direction
Magnitude and
Histogram Accum.
Convolution Mask
and shift matching
Restoration and
Retrieval
Spatial Error Block Analysis
14. Edge Direction & Error Block Classification
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos14
Edge Direction Classification
Uniform block: the gray level of
EB may be constant or nearly so.
I.e., there is no obvious edge in
the block.
Edge block: there are few edges
passing through the block and the
direction of each edge, in general,
is with no or little change.
Texture block: both gray level
and edge direction varies
significantly in the block, so the
edge magnitudes of many
directions are very strong.
Error Block Classification- 1
Histogram Accumulation
Error Block Classification- 2
15. Error Block Classification(2)
Bin Reduction:
The bin reduction of histogram of gradients is
used for classifying the edge blocks and texture
blocks.
It also can be used for improving the speed and
performance of our algorithm.
Bin: 59, 0,1, 14, 15, 16, 29, 30, 31, 44, 45 and 46
are most contributing for texture blocks
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos15
16. Block Rotation and Shape Formulation
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos16
Histogram
Characteristics
Convolution
Mask
Phase Offset
Calculation
Block Matching and
Shifting
17. Experimental Results(1)
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 17
Algorithm Pearson
Correlation
Spearman
Correlation
Block_msr -.721 .685
MGBIM [9] -.597 .584
S[63] .614 .570
Algorithm Pearson
Correlation
Spearman
Correlation
Block_msr -.843 .838
MGBIM [9] -.727 .925
S[63] .944 .937
Approaches Pearson Corr. Spearman Corr. RMSE
Wu and Yuen’s [9] .6344 .7365 7.1869
Vlachos’ [65] .5378 .7930 7.0183
Pan et al.’s [66] .6231 .6684 8.4497
Perra et al.’s [67] .6916 .6531 8.4357
Pan et al.’s [68] .5008 .6718 8.1979
Muijs & Kirenko’s
[69]
.7875 .6939 7.9394
Proposed Method .8627 .7104 7.0236
Pearson Correlation and Spearman for FUB database Pearson Correlation and Spearman for LIVE database
Test result using different approaches on the MPEG-2 video dataset
18. Experimental Results(2)
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 18
DATASET Wu et al.'s Pan et al.'s Mujis et al.'s Proposed
Sequence Recall Precision Recall Precision Recall Precision Recall Precision
LIV E : BlueSky 87.01 87.02 88.31 98.27 88.31 98.27 86.35 95.40
LIV E : Pedestrian 88.88 88.03 83.34 93.31 67.29 77.24 76.74 96.52
LIV E : RiverBed 76.58 86.50 87.57 97.57 64.28 74.89 75.54 92.26
LIV E : RushHour 77.64 87.54 86.83 96.83 68.80 78.02 77.63 90.60
LIV E : ParkRun 78.08 82.05 77.35 97.32 66.20 76.23 85.47 95.49
OCN : One 69.44 89.28 77.77 93.33 63.89 79.31 83.33 96.77
OCN : Mr:Big 70.23 88.67 79.41 90.94 68.56 75.42 85.58 98.11
OCN : Swim 66.87 85.72 84.56 95.24 65.55 78.56 88.23 95.46
Comparison of different algorithms showing the detection rate of distorted frames
19. Experimental Results(3)
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 19
Pattern orientation calculation considering histogram bin Bin Reduction : Selection of significant histogram bins
Rotation Formulation and Bin Reduction
Histogram Accumulation of Match and Shifting
20. Experimental Results(4)
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos 20
High Priority bins to take the decision:
32, (88, 90, 92) and 128
[For Matched Case, High accumulation].
Second high priority bins to take decision:
(14, 15, 16) and (44, 45, 46)
[Accumulation from High (full unmatched)
to zero (partially matched)].
Discussion and Decision
21. 21
We have proposed an efficient Video
artifact measurement and error frame
detection method- that does not
restrict itself only compression based
artifacts.
Major Contribution-1
Our Error block analysis algorithm is
less sensitive to illumination variation
and noise. Moreover, it can deal with
not only traditional artifacts but also
wireless transmission and broadcasting
related artifacts.
Major Contribution-2
Our analysis method can formulate the
distortion pattern rotation and shape-
in later part which can be used in video
restoration, concealment and retrieval.
Major Contribution-3
Feature Work
We will use the analytical parameters for video
error concealment. How we incorporate these
information for next step is a challenging
research issue.
Conclusion and Future Work
Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos
22. Publication List
SCI/SCIE Indexed Journals
1. Md. Mehedi Hasan, Kiok Ahn, Mahbub Murshed, Oksam Chae; “Hawkeye: A Cloud Architecture for Automated Video Error Detection in Real-time”,
INFORMATION Journal) (Accepted: 12th April, 2012) (SCIE) [ISSN: 1343-4500, E-ISSN: 1344-8994].
2. Md. Mehedi Hasan, Kiok Ahn, JeongHeon Lee, SM Zahid Ishraque, Oksam Chae; “Fast and Reliable Structure-Oriented Distortion Measure for Video
Processing”, Advanced Science Letters (Accepted: 6th December, 2011) (SCIE, IF: 1.253) [ISSN: 1936-6612, E-ISSN: 1936-7317].
International Journals
1. Md. Mehedi Hasan, Kiok Ahn, Oksam Chae; “Faster Detection of Independent Lossy Compressed Block Errors in Images and Videos”, International
Journal of Signal Processing, Image Processing and Pattern Recognition, vol. 5, no. 1,pp. 151-164, March, 2012)[ISSN: 2005-4254].
2. Md. Mehedi Hasan, Kiok Ahn, Oksam Chae; “Measuring Blockiness of Videos using Edge Enhancement Filtering ”, SIP, Communications in Computer
and Information Science (CCIS), vol. 260, pp. 10-19, January, 2012) (Springer- Verlag, Berlin-Heidelberg)[ ISSN: 1865-0929, ISBN: 978-3-642-
27182-3].
International Conference Papers
1. Md. Mehedi Hasan, Kiok Ahn, Md. Shariful Haque, Oksam Chae; “Blocking Artifact Detection by Analyzing the Distortions of Local Properties in
Images ”,ICCIT 2011, 14th International Conference on Computer and Information Technology, IEEE Xplore, pp. 475-480, Dec. 22-24, 2012) [ISBN:
978-1- 61284-907-2].
2. Md. Mehedi Hasan, Kiok Ahn, SM Zahid Ishraque, Oksam Chae; “Hawkeye: Real-time Video Error Detection Using Cloud Computing Platform ”,AIM
2012, Proceedings of the FTRA International Conference on Advanced IT, engineering and Management), pp. 121-122, Seoul, Korea, Feb. 6-8, 2012).
3. Md. Mehedi Hasan, Kiok Ahn, Oksam Chae; “Measuring Artifacts of Broadcasted Videos by Accumulating Edge Gradient Magnitude ”,YSEC 2012,
Proceedings of the 37th KIPS Spring Conference), Korea, April 26-28, 2012).
4. Md. Mehedi Hasan, Kiok Ahn, Mohammad Shoyaib, Oksam Chae; “Content- Based Error Detection and Concealment for Video Transmission over
WLANS ”,AIM Summer 2012, Proceedings of the FTRA International Conference on Advanced IT, engineering and Management), Jeju, Korea, July 10-
12, 2012) [Accepted].
5. Mahbub Murshed, SM Zahid Ishraque, Md. Mehedi Hasan, Oksam Chae; “Cloud Architecture for Lossless Image Compression by Efficient Bit-Plane
Similarity Coding ”, AIM 2012, Proceedings of the FTRA International Conference on Advanced IT, engineering and Management), pp. 123-124,
Seoul, Korea, Feb. 6-8, 2012).
6. Minsun Park, Md. Mehedi Hasan, Jaemyun Kim, Oksam Chae; “Hand Detection and Tracking Using Depth and Color Information ”,IPCV 2012, The
2012 International Conference on Image Processing, Computer Vision, and Pattern Recognition), Las Vegas, USA, July 16-19, 2012).
22Edge-Based Feature Extraction for Artifacts Detection and Error Pattern Analysis from Broadcasted Videos
Existing methods fail due to the difficulty to manage motion in the background.
Existing moving object detectors fail when motion-free backgrounds are not available.
Existing segmentation methods cannot separate them.
Existing shape matching methods can not track shape and color variation at the same time.