Lossless compression of mammographic images with region-based predictor selection

Abstract

One of the main tools for early diagnosis of breast cancer is digital mammography. These images require large storage space and are difficult to be transmitted over communication links. In this paper we propose a context-based method for lossless compression of these images. Some modifications are performed to customize the activity level classification model (ALCM) predictor to work best in mammograms. The function of the modified predictor changes for different main regions of these images. Also, best qualities of two other predictors are exploited and the results of the fittest predictor are selected adaptively for the prediction of a pixel. Moreover, context modeling is used for a better categorization of the prediction errors. The proposed algorithm was tested using images from a well-known database and the results were compared with two standard compression methods of lossless mode of JPEG2000 and JPEG-LS. The proposed method was proved to produce better compression results than those of the standard algorithms. © 2013 Institute of Electrical Engineers of Japan.