Two-Phase Compression of Histological Images with MDL Ranking of Segmentation Images

We propose several methods for minimizing the codelength achievable when encoding histological images in two phases, where a segmentation of the image is encoded in the first phase and the image conditional on its segmentation is encoded in the second phase. The main goal of the paper is to establish the class of models suitable to be used in minimum description based analysis of histological images, and to compare the compression performance achievable by several predictive coding schemes, which we derive specifically for the second phase. The most efficient of the publicly available general image compressors, JPEG-LS and CALIC, are also compared in the one phase coding scenario. We conclude with a preliminary study on setting automatically the optimal parameters of one segmentation algorithm so that the compressed description has minimum length.