Digital Dropout detection in degraded media archives
Md. Monirul Hoque
Digital dropout DCT HVS SVM Edge detection;
- 원문 URL
With the rapid development of visual digital media, the demand for better quality of service has increased the pressure on broadcasters to automate their error detection and restoration activities for preserving their archives. Archiving contents are typically degraded due to physical problems in repeated projection or playback or simply the chemical decomposition of the original material. In order to preserve the archive materials by converting these degraded contents to digital file, it is possible that noise or errors contained in the film or tape is either maintained or displayed in other forms. In many video processing applications, accurate knowledge of these errors or distortions present in the input video sequence is very important. These errors and noise reduce the quality of the generated files. As the amount of data is quite large, manual retouching is unfeasible. Hence, the area of automated error detection of image sequences has gained significant industrial attention. Digital dropout is one of the defects that affects archived visual materials and tends to occur in block by block basis (size of 8X8). Digital dropouts are a major type of damage when storing digital AV content on digital video tape carriers or when transferring this content to file based environments. Despite the practical importance of algorithms for the detection of dropouts originating from digital video tapes (Digi Beta, IMX etc.), there is only very little related scientific work focusing on image based methods focusing on this kind of defect. There exist a number of mechanisms to detect transmission/compression related error in H.263++/H.264++ codec but the hypothesis and features used in these methods are tailored for erroneous macroblocks (MBs)/group-of-blocks(GOBs) detection. So those methods performance for digital dropout detection are likely be poor. Another problem of the spatio-temporal features used in those methods is that, in presence of pathological motion, motion estimation will fail i.e. in some parts of any sequence it will be impossible to model the behavior. So coping with pathological motion remains an unsolved issue for those approaches. In this dissertation, we focus towards the development of digital dropout detection algorithm. To address the above issues, we propose two computationally convenient approaches based on features related to human visual perception and relevant DCT coefficients respectively. We also propose an optimal weighted neighborhood sampling strategy based on block dominant direction to enhance the discriminative ability of block representation. Both of the proposed methods work based on spatial frame; hence works independently in presence of fast/pathological motion. It is well established that human visual system (HVS) is highly adapted to the statistics of its visual natural environment. Consequently, in our first method, we have formulated digital dropout detection as a classification problem which predicts block label based on block statistical features. These block statistical features are indicative of perceptual quality relevant to human visual perception, can distinguish pristine image blocks from distorted ones. In the second method, we first select the discriminant DCT coefficient set through a hybrid feature selection method. Here, the idea is to learn relevant DCT coefficients which can separate error blocks from normal ones. In both methods, we have introduced an optimal weighted neighborhood sampling strategy to enhance the discriminative ability of block representation. In the final stage of our methods, a SVM classifier is used to detect the digital dropout error block in the video frames. Finally, experiments on real video archive show that the proposed algorithms are efficient to detect digital dropout error with minimal false detection.This ensures that no loss in quality is experienced when the video is in error-free condition. We also show that both method correlates highly with human subjective judgments of quality.