Automatic Video Error Detection and Restoration for Broadcasting and Archiving System

다국어 초록 (Multilingual Abstract)

In recent years, there has been a growing interest in the old archived videos owing to having significant cultural and historical records. Most of the old archive-management organizations do not consider the quality of videos, rather aim only to archive the videos in large numbers. Contrariwise, the recent advancement in the amenity of digital videos including the wider screen, better resolution, and screen-contrast has made the users habituated with experiencing better-quality videos. Hence, to address this issue, most archive-management organizations convert the old digital/analog media into a new digital file before archiving. This digitization process may introduce new errors in the file and/or inherit the errors in the original analog tape, causing degradation to the quality of the video. Existing efforts have tried to tackle this issue through human visual inspection, which at times suffers from efficiency due to the large number of video contents. To overcome such limitations of human visual inspection, automated quality control (QC) has been proposed in the literature, which also shows its shortcomings in error detection, error recovery, and system implementation processes. More specifically, the error-detection process may suffer from either high false discovery rate(FDR), or inability to capture errors at the same position for consecutive frames having a different format. The error recovery process undergoes with blurred-boundary in some videos having large motion variations. Moreover, most existing systems are not designed to address each possible error, and most importantly, they do not support real-time inspection of large video archives.

To solve these problems, this work aims at designing an automatic video-error detection and restoration system, broadly covering three major sections including error-detection, error-recovery, and system implementation. In particular, we perform digital dropout detection through introducing a discriminant DCT coefficient-driven block descriptor. We also propose a blotch-detection algorithm based on five-frame extended Rank Order Detector(ROD) to detect the errors occurring in the same position of consecutive frames. In the case of error-recovery, we propose a method that measures the reliability of each pixel in the error frame followed by reconstructing it from the surrounding pixels on the space-time domain by considering motion estimation. Besides, we propose a method for restoring color degradation, using inverse transformation process by formulating a color degradation model, based on degraded color bar as polynomials. Finally, we offer an inspection system that integrates the video above mentioned error detection/restoration techniques. We also propose a cloud-based system for automatic synchronization of error detection and error recovery systems, which is especially useful for large-capacity video archiving.

This system is commercialized as a product called Hawkeye that supports the detection of more than twenty errors. However, in this dissertation, we deal with some specific errors that occur highly in most archived media. The proposed system has proven its performance in the real broadcasting station and archive environments.

목차 (Table of Contents)

• 1 Introduction 1
• 1.1 Overview 1
• 1.2 Existing automatic video error detection and restoration systems 2
• 1.3 Major Contribution 4
• 1.4 Organization of the dissertation 6
• 2 Related research and Background study 8
• 2.1 Literature review 8
• 2.1.1 Motivation 9
• 2.1.2 Video error classication 10
• 2.2 Error detection: Area of Interest 21
• 2.2.1 Digital dropout 21
• 2.2.2 Blotch 23
• 2.2.3 Betacam dropout 24
• 2.3 Error Restoration: Area of Interest 25
• 2.3.1 Analogue tape dropout 27
• 2.3.2 Color degradation 28
• 2.4 Summary 30
• 3 Video Error Detection 32
• 3.1 Digital dropout 33
• 3.1.1 Problem overview 33
• 3.1.2 Coecient selection 36
• 3.1.2.1 Initial coecient set 36
• 3.1.2.2 Genetic algorithms 37
• 3.1.3 Feature representation 40
• 3.1.4 Support Vector Machine Classier 42
• 3.2 Blotch 44
• 3.3 Betacam dropout 48
• 4 Video Error restoration 51
• 4.1 Digital dropout 53
• 4.2 Black-frame error 54
• 4.3 Analogue tape dropout 55
• 4.4 Color-Degradation 58
• 4.4.1 Color Transformation Polynomial 58
• 4.4.2 Converting Polynomial Solution Using Color bar 59
• 4.4.3 Restoring Degraded Color without Color bar 61
• 5 System Implementation 62
• 5.1 System overview 63
• 5.2 Real-time instant processing 67
• 5.3 Cloud based video error detection system 72
• 5.4 Automated video restoration system 75
• 6 Experiment and Analysis 79
• 6.1 Video error detection 79
• 6.1.1 Digital dropout 79
• 6.1.1.1 Database Description 79
• 6.1.1.2 Classication Measures: Quantitative Analysis 80
• 6.1.1.3 Classication Measures: Human subjective qualityrating analysis 82
• 6.1.1.4 Experimental Results: Feature Selection Analysis 82
• 6.1.1.5 Experimental Results: Proposed Algorithm Analysis 89
• 6.1.1.6 Experimental Results: Human subjective quality ratinganalysis 91
• 6.1.1.7 Experimental Results: Dataset overtting analysis 91
• 6.1.1.8 Comparison with other approaches 92
• 6.1.1.9 FDR measurement in archive environment 93
• 6.1.2 Blotch 94
• 6.1.3 Betacam dropout 96
• 6.2 Video Error Restoration 101
• 6.2.1 Digital dropout 101
• 6.2.2 Analogue tape dropout 102
• 6.2.3 Black frame 102
• 6.2.4 Color degradation 106
• 6.2.5 Similarity between frames by reconstruction using PSNR 109
• 6.3 System Implementation 110
• 6.3.1 Real-time instant processing 110
• 6.3.2 Cloud based video error detection system 114
• 6.3.3 Automated video restoration system 116
• 6.3.4 Summary 118
• 7 Conclusion 120
• Bibliography 124
• A List of Publications 134

참고문헌 (Reference)

1. 국가기록원 소장 영화필름의 디지털 복원 필요성에 관한조사,, 김상국, 유병문, 정경택, 기록물 보존복원, vol.4,pp.43-54, (cit. on p. 3), , 2011

2. 코너 정보와 Adaboost 인식기술을 이용한 비디오 내의 블록 오류 고속 검출 방법,, 안기옥, 이문식, 하명환, 박성춘, 김민기, 한국방송공학회, in 한국 방송공학회 학술발표대회논문집, , pp. 58-61 (cit. on pp. 27, 94), , 2011