In this paper we propose a novel quadtree decomposition coding algorithm for residual frames in video coding. First we employ overlapped block motion compensation to reduce blocking artifacts

this leads to smoother motion prediction frames which helps to encode the residual frames more efficiently; based on the nature of human visual system

we regard each 2×2 subblock as a basic coding unit

compute its mean value

and use a new quadtree decomposition method to label the subblocks with larger magnitude than a chosen threshold

then code them with scalar quantizer and adaptive arithmetic coder. Compared with transform-based methods

such as EZW and H.263

the proposed algorithm is more efficient and has better performance

which indicates that it is a very effective and computationally simple technique especially for low bit rate coding. Moreover

the coding and decoding procedures are so fast that it can satisfy real-time applications.