DEADZONE QUANTIZATION METHOD AND APPARATUS FOR IMAGE COMPRESSION

A method for efficiently optimizing the bin widths for a distribution or an image to be compressed. An image having symmetric uni-modal distribution is divided into a zero bin having a zero bin width and a plurality of outer bins having an outer bin width. M numbers of predetermined candidate values for the zero bin width and N numbers of predetermined candidate values for the outer bin width are provided. A zero bin probability is derived from an entropy function. The allowable zero bin width is calculated from the zero bin probability and target bit rate. The allowable zero bin width is then searched to obtain an optimum combination of the zero bin width and the outer bin width, the optimum combination being the combination having the least distortion measure. In one embodiment, a fast algorithm is used to search the outer bin width for each given zero bin width, reducing the required combination by a factor of logN/N. In one embodiment, the allowable zero bin width is a subset of M number of candidate values for the zero bin width. In one embodiment, the middle number of an interval is used as the quantization value. In another embodiment, the centroid of an interval is used as the quantization value.