VIDEO CODING USING WAVELET BASED CONTOUR AND TEXTURE EXTRACTION

Abstract

For very low bit-rate video coding (64 kb/s and below), the major technological challenge resides in the development of suitable compression algorithms that can both achieve the desired bit-rate and permit real-time implementation at economical costs. Multiscale edge detection and reconstruction algorithm was proposed for edge extraction and reconstruction in Differential Frame Differences (DFD) signals. The algorithm overcomes the conventional problem that the reconstructed signal has obvious blocking effects when using traditional Discrete Cosine Transform (DCT) encoding approaches. Preliminary investigation was performed to facilitate the understanding of the signal characteristics of the DFD signals and the suitability of various coding approaches. After extracting the texture from the wavelet analysis, DCT was adopted to compress the smoothed texture. In order to reduce the bit rate, texture was subsampled before DCT was carried out without introducing too much degradation in the final reconstructed image quality. After a discussion of texture coding schemes, the contour coding schemes were introduced. The multiscale edge detection and reconstruction algorithm maintains most of the original signals' energy in the high frequency subbands and helps to substantially reduce the edge points. Motion compensation and subsampling provided the first level of significant information reduction. Efficient coding of active coded blocks using multiscale edge detection and reconstruction completed the design of the video coder. The system performance was analyzed with the help of the collected data in the corresponding three methods in high frequency subband coding. On the basis of the three methods of chain selection and the three methods of subsampling, the integrated system was built. The integrated scheme was shown to be more suitable for low bit rate video coding application with satisfactory compression performance and reasonable quality. It can obtain Peak Signal-to-Noise Ratio (PSNR) in the range of 32.80dB to 33.80dB when its bit rate is in the range of 0.03bpp to 0.06bpp