VARIABLE-RATE VECTOR QUANTIZATION FOR SPEECH, IMAGE, AND VIDEO COMPRESSION

The performance of a vector quantizer can be improved by using a varia ble rate code. We apply three variable rate vector quantization system s to speech, image, and video sources and compare them to standard vec tor quantization and noiseless variable rate coding approaches. The sy stems range from a simple and flexible tree-based vector quantizer to a high performance, but complex, jointly optimized vector quantizer an d noiseless code. The systems provide significant performance improvem ents for subband speech coding, predictive image coding, and motion co mpensated video, but provide only marginal improvements for vector qua ntization of linear predictive coefficients in speech and direct vecto r quantization of images. We suggest criteria for determining when var iable rate vector quantization may provide significant performance imp rovement over standard approaches.