Learning discriminant DCT coefficients driven block descriptor for digital dropout detection system in degraded archived media

• Identify a set of DCT coefficients that can be used in digital dropout error classification.
• A weighted neighborhood sampling strategy based on spatially correlated directional behavior.
• Feature extraction in DCT domain, resulting in lower time complexity and computational load.
• Correlates highly with human subjective judgments of quality of error.

Digitization of old archived media is of great importance to preserve the originality of medium in terms of historical record as well as the means to quality improvement for reproduction purposes. However, digitization increases the exposure of the media to digital dropout error, thus presenting a significant degradation in perceptual quality of the converted video sequences. A numbers of mechanisms were investigated in the past to make these converted media more robust against digital dropout errors. Nevertheless, these techniques achieved little success, forcing manual quality check to assure standard quality. This paper presents an automatic solution to this problem based on discriminant DCT coefficients. Here, the idea is to build a block classification model by learning discriminant DCT coefficients first and utilize these coefficients along with an weighted neighborhood sampling strategy to formulate discriminant block descriptor so that within-class difference of the block features is minimized and between-class difference is maximized. This spatial detection is free from motion computation; thus performs accurately in presence of pathological motion (PM) and fast moving objects. Finally, the proposed method is compared against the existing methods to demonstrate improved detection accuracy using real degraded video archives.