Low-bit-rate image coding using sketch image and JBIG

At low bit rates most current image coding techniques will generate a smooth image where some of the important details of the original image are lost. For example, the compression of an image by JPEG down to a bit rate of 0.1 to 0.2 bit/pel usually results in a significant loss of detail and an annoying blocking effect. The basic idea in this paper is that in applications where the detail information in a low bit rate coded image is of more concern than the gray scale information, it is necessary to encode the former and disregard the latter. Based on the logarithmic image processing model, we have recently proposed a novel algorithm to generate a binary image (called the sketch image) from a gray scale image. We have shown that the sketch image retains much more information of the original image than those images generated by other image binarization techniques such as simple thresholding, ordered dither and error diffusion. In this paper, we further study the sketch image algorithm and propose a simple data compression scheme which incorporates a decimation and interpolation algorithm and the JBIG algorithm. Experimental results have shown that at low bit rate (about 0.1 bit/pel) the proposed algorithm preserves the fine details of the original image at the expense of loss of grey scale information.