FUZZY MOTION ESTIMATION AND COMPENSATION FOR VIDEO COMPRESSION = 애매논리를 이용한 영상압축을 위한 움직임 예측 및 보상

다국어 초록 (Multilingual Abstract)

This dissertation shows how fuzzy systems can help compress image sequences. The fuzzy systems filter noise and improve motion estimation and compensation. Adaptive fuzzy systems further improve the compensation accuracy. The dissertation describes three applications of fuzzy imagesequence compression: fuzzy filters for impulsive noise, fuzzy virtual tilings for subband image coding, and fuzzy motion estimation and compensation.
Fuzzy systems of if-then rules can filter impulsive noise from signals. An additive fuzzy system learns ellipsoidal fuzzy rule patches from a new pseudo-covariation measure of alpha-stable covariation. Mahalanobis distance gives a joint set function for the learned if-part fuzzy sets of the if-then rules. The joint set function preserves input correlations that standard factored set functions ignore. The fuzzy system filtered such noise better than did a benchmark radial basis neural network.
A fuzzy system can improve how subband coding compresses images. A fuzzy system maps the coefficients of frequency subbands to virtual tiles of the time-frequency plane. This prunes the tiling tree and gives a high-energy subtree. The new tiling subtree shapes the error spectrum and helps preserve edges in the image.
A fuzzy system can also estimate motion vectors and increase the compensation accuracy for video compression. The fuzzy system uses the temporal correlation of the motion field to estimate the motion vectors. First and second order statistics of the motion vectors give ellipsoidal search windows. Out algorithm reduced the search area and gave clustered motion fields. We also proposed a fuzzy overlapped block motion compensator. The fuzzy system uses the motion vectors of neighboring blocks to map the previous frame's pixel values to the current pixel value. The rules come from the previously decoded frame. The fuzzy system updates its rules as it decodes the image. The fuzzy system also improved the compensation accuracy.

목차 (Table of Contents)

• Abstract = xviii
• 1 Introduction = 1
• 1.1 Video Compression Systems = 1
• 1.2 Current State of Research = 3
• 1.3 Dissertation Objective = 4
• 1.4 Dissertation Outline = 5
• 2 Additive Fuzzy Systems = 6
• 2.1 Additive Fuzzy Systems = 6
• 2.2 SAM Theorem = 8
• 2.3 Learning in SAMs: Unsupervised Clustering and Supervised Gradient Descent = 12
• 3 Fuzzy Prediction and Filtering in Impulsive Noise = 18
• 3.1 Abstract = 18
• 3.2 Filtering Impulsive Noise = 19
• 3.3 Alpha-stable Noise: Covariation Versus Covariance = 21
• 3.3.1 Stable Distribution and Its Properties = 22
• 3.3.2 Covariation and Pseudo-covariation = 23
• 3.4 Covariation Rules and Mahalanobis Sets: Product Space Clustering and Projection = 26
• 3.5 fizzy Prediction with Covariation Rules = 31
• 3.6 Fuzzy Filters with Covariation Rules = 34
• 3.7 Conclusion = 39
• 4 Fuzzy Virtual Tilings for Subband Image Coding = 46
• 4.1 Abstract = 46
• 4.2 Things of the Time-Frequency Plane = 47
• 4.3 Subband Image Coding = 49
• 4.3.1 Subband Filters in Two Dimensions = 49
• 4.3.2 Quadrature Mirror Filters = 50
• 4.3.3 Fully Decomposed SBC of Images with Vector Quantization = 54
• 4.4 Fuzzy Virtual Things for Subband Coding = 56
• 4.5 Simulation Results = 67
• 4.6 Conclusion = 72
• 5 Fuzzy Motion Estimation and Compensation
• 5.1 Abstract = 73
• 5.2 MPEG Standards for Video Compression = 74
• 5.3 Motion Estimation and Compensation = 75
• 5.3.1 Motion Estimation = 75
• 5.3.2 Motion Compensation = 78
• 5.4 Motion Estimation Using Adaptive Vector Quantization = 79
• 5.4.1 Competitive AV(a Algorithm for Local Means and Covariances = 79
• 5.4.2 Motion Estimation Using Adaptive Vector Quantization = 83
• 5.4.3 Unsupervised Learning for AVQ Algorithm = 84
• 5.4.4 Complexity Analysis = 85
• 5.5 Fuzzy Overlapped Block Motion Compensation = 86
• 5.5.1 Overlapped Block Motion Compensation = 87
• 5.5.2 Fuzzy Overlapped Block Motion Compensation = 89
• 5.6 Simulation Results = 93
• 5.6.1 Motion Estimation = 93
• 5.6.2 Motion Compensation = 99
• 5.7 Conclusion = 102
• 6 Conclusions and Future Research = 103
• 6.1 Conclusions = 103
• 6.2 Future Research = 104
• Bibliography = 120
• A Fuzzy Throttle and Brake Control for Platoons of Smart Cars = 115
• A.1 Abstract = 115
• A.2 Introduction = 116
• A.3 Learning Fuzzy Rules = 118
• A.3.1 Unsupervised Rule Estimation with Competitive Learning = 118
• A.3.2 Supervised Ellipsoidal Learning = 120
• A.4 Controller Structure = 123
• A.4.1 Velocity Controller = 123
• A.4.2 Gap Controller = 124
• A.4.3 Brake Controller = 129
• A.5 Car Models and Sensor System = 134
• A.5.1 Car Models = 135
• A.5.2 Brake Models = 136
• A.6 Simulation and Test Results = 137
• A.6.1 Learning Fuzzy Rules = 137
• A.6.2 Gap Controller with Throttle Only = 138
• A.6.3 Roadway Tests for Throttle Controller = 139
• A.6.4 Gap Controller with Throttle and Brake = 144
• A.7 Conclusion = 153