Adaptive transform coding with multi-band vector quantization for video signals

A hybrid video coding algorithm based on the combination of motion compensation, the discrete cosine transform, and a bank of adaptive N-dimensional vector quantizers (AVQ) is described. The motion estimation and compensation are performed in the pixel domain on a block-by-block basis. Only the blocks with significant prediction errors are transformed and vector quantized. The blocks are decomposed into bands with variable width using zig-zag ordering. Adaptivity to input signal statistics is achieved by identification of the prediction errors and various AC-energy thresholds, which are used as keys to classify the blocks into four types and assign bits to them adaptively from frame to frame. The results of application of this image compression technique to test images are presented, showing good performance at 128 and 256 kb/s.<<ETX>>