Near-lossless image compression by combining wavelets and CALIC

In this paper we report experimental work on L/sub /spl infin//-constrained high-fidelity image compression. In our experiments minimum-entropy pre-quantization and minimum-entropy trellis quantization did not lead to competitive L/sub /spl infin//-constrained near-lossless compression performance against our previous work of L/sub /spl infin//-constrained CALIC (Wu et al. 1997). But we were able to improve L/sub /spl infin//-constrained CALIC, the best L/sub /spl infin//-constrained image coder for error tolerance below 7 so far by a hybrid approach. First, an input image is wavelet transformed, the wavelet coefficients are quantized in an L/sub /spl infin// criterion and entropy coded. Then a wavelet approximation of the image is constructed by inverse transform based on the quantized wavelet coefficients. Finally, L/sub /spl infin//-constrained CALIC is used to compress the residue image between the wavelet approximation and the original.