Compressing High Bit Depth Images of Sparse Histograms

To improve the lossless compression ratios for images having sparse histograms, a method of histogram packing was introduced. The method was found to be effective for low bit depth images. We investigate effects of packing histograms of high bit depth images—medical CR, CT, and MR images as well as various natural 16-bit ones. We analyze an off-line packing method, which requires encoding the original histogram along with the compressed image. We present several methods of histogram encoding and analyze their usefulness. One of them (RLE+LZ77) obtains the shortest encoded histogram length for nearly all tested images and in practice is sufficiently good for encoding histograms of wide range of images. A simpler method (MT) may be useful for medical images. For these images, its use results in improvements of the compression ratio little worse compared to RLE+LZ77, but decoding of images with histograms encoded using the MT method is already supported by JPEG-LS and JPEG2000 (part 2) standards. Effects of histogram packing are examined for the CALIC, JPEG2000, and JPEG-LS algorithms. Histogram packing improves significantly lossless compression ratios for high bit depth sparse histogram images. The ratio improvement may exceed a factor of two, as in the case of MR medical images.