Embedded Image Coding Based on Laplacian Pyramids with Quantization Feedback

In this paper, a multi-layer SNR-scalable error-bounded image encoder is achieved in the framework of Laplacian pyramids with quantization noise feedback, by exploiting an entropy-minimizing optimum quantization strategy, a content-driven decision rule based on an L-infinity activity measure, and multistage quantizers to progressively upgrade quality at full scale. The resulting scheme yields intermediate versions with scale and SNR both increasing, and a further SNR scalability on the full resolution, with possibly lossless reconstruction, thereby expediting interactive browsing of remote data bases of images of any sizes and wordlength. The proposed encoder outperforms JPEG which does not possess all the above mentioned attractive characteristics.