On content-based very low bitrate video coding

Abstract

A new approach to video coding based on content activity of sequences is introduced. We use a three-dimensional generalization of the Discrete Cosine Transform (3D-DCT) to construct a coding system. Exploiting both spatial and temporal redundancies of video sequences, a very-low bit-rate encoder is developed and analyzed. The human visual system (HVS) characteristics are used to further reduce the bit-rate, achieving a compression ratio of more than 150:1 with good quality of the reconstructed sequence, measured as more than 37 db (PSNR). The complexity of the proposed approach is significantly lower than existing systems such as H.263, making it suitable for mobile applications. © Springer-Verlag Berlin Heidelberg 2003.