Binary tree-structured decomposition for lossless medical image compression

Abstract

A lossless image compression is important for medical image since any information loss or error during image compression process can affect clinical diagnostic decision. This thesis proposes a lossless compression algorithm for major application to medical image which has high spatial correlation. The proposed image compression algorithm has binary tree-structured decomposition scheme in conjunction with prediction and classification. In the proposed algorithm, an image is divided into four subimages by subsampling, one of which is used as a reference subimage to predict three other subimages. The prediction error of three subimages is classified into two subsets based on a slope change of the reference subimage, and the classified errors are encoded by entropy coding with corresponding codewords, respectively. This subsampling(decomposition) and classified entropy coding processes are repeated to the reference subimage, and the last reference subimage is encoded by conventional differential pulse code modulation(DPCM) and entropy coding. In order to verify this proposed algorithm, it is applied to several chest X-ray, X-ray CT, and MR images, and the results are compared to the well-known lossless compression algorithms.