Compressing images of sparse histograms

Most single-frame single-band medical images, like CR, CT, and MR, are of a high nominal bit depth, which usually varies from 12 to 16 bits per pixel. The actual number of pixels' intensity levels found in those images may be smaller, than implied by the nominal bit depth, by an order of magnitude or even more. Furthermore, levels are distributed throughout almost all the entire nominal intensity range, i.e., the images have sparse histograms of intensity levels. Image compression algorithms are based on sophisticated assumptions as to characteristics of the images they process. Sparse histogram is clearly different from what is expected by lossless image compression algorithm, both in case of predictive and of transform coding. To improve the compression ratios of such images, a method of histogram packing was recently introduced. The method is found to be effective, however, the research was done for low bit depth images. In this paper, we investigate effects of packing histograms of high bit depth medical images. We analyze an off-line packing method and find it to be highly effective. The off-line packing requires the information, describing how to expand the histogram after decompressing an image, to be encoded along with the compressed image. We present an efficient method of encoding this information. Experiments are performed for CALIC, JPEG2000, and JPEG-LS. The effects of packing histograms on the compression ratios of tested algorithms are, for all the tested algorithms, very similar. The average compression ratio improvement obtained for the CR, CT, and MR images is about 15%, 42%, and 52% respectively.