Wavelet Based Compression of Digital Mammograms

The requirements for large amounts of medical imaging data that have to be stored eeciently and transmitted at low bit rates, has generated considerable interest in the design of eeective compression algorithms. The main challenge in medical image compression is to achieve high compression ratios while preserving important diagnostic information. The current Joint Photographic Experts' Group (JPEG) standard for still image compression is based on the block Discrete Cosine Transform (DCT). As a result at low bit rates blocking artifacts occur at the borders of the subblocks into which the original image is partitioned. In the case of medical images these artifacts, apart of being visually annoying, can also compromise diagnostic accuracy. Subband based coding schemes have been receiving increased interest in the image compression eld as they outperform JPEG and overcome the blocking eeect problem at low bit rates. The Wavelet Transform (WT) 1] has gained wide application in the design of octave band decomposition systems. A variety of methods have been proposed in the literature for the encoding of the wavelet coeecients. A hierarchical scheme that exploits the similarity across subbands of the same orientation is the embedded zero tree wavelet algorithm (EZW) of Shapiro 2]. The employment of Vector Quantization (VQ) 3] in the wavelet transform domain has also been investigated. Examples of WT-VQ based coding schemes are the lattice VQ of 4] and the entropy constrained VQ of 5]. The application of wavelets in the compression of digital mammograms has been reported in 6] and 7]. In this paper we present the application of a novel wavelet based still image codec to the compression of digital mammograms. An adaptive VQ scheme has been developed for the quantization of the wavelet coeecients. The available bit budget is optimally distributed among the subbands and noise shaping is performed according to the characteristics of the Human Visual System (HVS). The ability of the proposed system for progressive transmission (PT) of images is demonstrated. Progressive transmission is important in image browsing applications where it is desirable to allow the receiver to access a recognizable image as quickly as possible at minimum cost. Then, it can be decided whether to receive further image details or abort transmission. The Wavelet Transform is localized in both space and time and produces a multiscale representation of a function. The basic idea is to decompose any arbitrary function f using a set of basis functions …