Lossless compression of medical images using hierarchical autoregressive models

This paper introduces new hierarchical autoregressive (HMAR) models for lossless compression of medical images. The proposed concept involves a multi-layered modeling approach. The 2-D HMAR models can be thought of as modified versions of one-dimensional hierarchical AR (1-D HAR) signal models. The first layer of a 1-D HAR consists of a conventional AR model for the data, whereas each subsequent layer, in turn, attempts to model the AR coefficients of the preceding layer using a new AR model. The main advantage of HMAR is that the transmission of blockwise model coefficients becomes unnecessary. The performances of the proposed technique is compared with two existing alternative techniques; namely, hierarchical interpolation (HINT), and fixed differential pulse code modulation (DPCM). In terms of compression efficiency, HMAR models performs better than the other techniques considered.