Adaptive entropy-coded 2-D DPCM encoding of images

The authors describe the theory and performance of a particular adaptive entropy-coded quantization (AECQ) scheme used in conjunction with 2-D DPCM (differential pulse code modulation) and operating on real-world image sources. The resulting system, called 2-D DPCM/AECQ, is described in detail and its performance illustrated on a typical real-world image. The system makes use of an adaptive buffer-control strategy to modify the DPCM quantizer characteristics on the basis of the buffer occupancy state. This adaptive buffer-state feedback is used to control the average rate of the variable-length entropy-coded quantizer code words at the input to the buffer, to allow interfacing to a fixed-rate transmission or storage channel. Image coding results demonstrate, both objectively and subjectively, that achievable performance is close to that predicted on the basis of rate-distortion considerations even for reasonably small buffer sizes.<>