Using Multiscale Edge Detection For Low Bit-Rate Subband Video Coding

Abstract

This paper introduces a video compression system for very low bit-rate videoconferencing and tele-monitoring applications. The video codec first performs a three-dimensional subband decomposition on a group of video frames, and then encode the subbands with different quantization methods. By using a cubic spline wavelet, the spatial filter bank acts as a multiscale edge detector. This property is used for efficient selection and coding of high frequency subbands with geometric vector quantization. For lowest tempo-spatial subband, a DPCM coding with an entropy coder was used. Results at several low bit rates (16, 32, 64, 128 Kbps) are reported and compared with H.263, the standard for low bit rate video coding.