Adaptive Predictive MAR Models for the Medical Image Compression

• Main Authors: Chen, Zuo-Dian, 陳作典
• Other Authors: Ruey-Feng Chang
• Format: Others
• Language: zh-TW
• Published: 1996
• Online Access: http://ndltd.ncl.edu.tw/handle/05502153443002107650

碩士 === 國立中正大學 === 資訊工程學系 === 84 === ABSTRACT In the thesis, the adaptive predictive multiplicative autoregressive (APMAR) method and progressive APMAR (PAPMAR) are proposed and implemented in the lossless medical image coding. We improve the prediction accuracy for the blocks which need to be encoded by using adaptive predictor to process the image in advance. The predictive accuracy for a block of the method is better than the fixed prediction method. We use seven predictors of the JPEG lossless mode and a local mean predictor which has been implement in the space- varying multiplicative autoregressive (SMAR) to predict the blocks, then use MAR to process the residual values. The final residual values were processed by Huffman coding. PAPMAR has two passes. The important information which is useful for the diagnosis is encoded in the first pass of PAPMAR and the residual information is encoded in the second pass of PAPMAR. The purpose of our these methods is to obtain a higher compression ratio for the medical images. The first step of APMAR and PAPMAR is to split an image into blocks. In the APMAR compression method, the correlation of pixels is first reduced by the first predictive process and the correlation of the residual pixel values is further reduced by the second predictive process. APMAR and the progressive coding are implemented in PAPMAR. In the first pass of PAPMAR compression method, an image is separated into regions of interest (ROI) blocks and non-ROI blocks which are the blocks except ROI blocks in the image. Only the ROI blocks are encoded by the APMAR method in the first pass of PAPMAR and the non-ROI blocks are filled with the background pixel value. In the second pass of PAPMAR, non-ROI blocks are subdivided into true non-ROI blocks which have the same pixel values with the background pixel value and pseudo non-ROI blocks which have similar but not the same pixel values with the background pixel value. Only the pseudo non-ROI blocks are encoded in the second pass of PAPMAR. The encoded image quality of the first pass of PAPMAR is very high especially for the diagnostic part because the ROI blocks are encoded by the lossless compression method. The new improved encoding scheme was tested on several different medical images, such as a set of brain and knee MR images, chest CT images, heart UT images, chest X-ray images, etc. The results are compared with those typical and famous medical image compression methods, such as the MAR, SMAR, DPCM, and HINT schemes. The new method was more applicable to the general medical images.