Motion Oriented Picture Interpolation with Human Perceptual Consideration

Abstract

The demand for bit rate reduction in video signals for future applications (e.g., transmission via broadband ISDN) is getting stronger. To achieve such an efficient coding performance, the field or frame skipping technique can be used. In this paper, we develop a new picture interpolation system which classifies every block of the interpolated picture into different motion types and then applies quality preserving methods to encode each type of block. The system first detects still blocks by an examination of the available motion information from the anchor frames. If a block is not still, a hierarchical search algorithm is then utilized to classify each block into still, low motion, or high motion blocks and each block is then coded by a suitable technique to ensure that the reconstructed picture's quality is high. In addition, the luminance sensitivities of the human visual system are taken into consideration to adaptively adjust both thresholds so that subjectively good quality pictures can be obtained. A dynamic mapping variable-length code (VLC) is also proposed to more efficiently encode part of the transmitted data more efficiently than the fixed VLC. Its coding principle is to dynamically change the mapping relationship between the symbols and their corresponding codewords according to the present codeword statistics while keeping all codewords unchanged. Compared with the "B-picture" coding technique in MPEG, the proposed system shows a saving of more than 40% in transmitted data, as well as requiring fewer computations than MPEG.