An analysis on hybrid transform models for image compression in telemedicine applications

Abstract

As the field of medical imaging has attained a greater pace of development; the processes of medical imaging, its analysis and compression are ever competitive and remain as an ever-challenging demand especially in the field of telemedicine. The modalities of medical imaging range from various types of sources, and such medical images are required to be efficiently stored and communicated through the network from one place to other for diagnosis and other purposes. So, medical image compression is an important and vital requirement as it decreases the storage space and that in turn reduces the cost of network bandwidth. Further, for achieving better compression ratio and quality-based decoding as well as preserving the essential information in the medical image is a non-trivial task in medical image compression. In order to contemplate the requirements in medical image compression methodologies, many advanced compression models have been proposed in the literature. The works so far carried in the field of medical image compression have largely focused on addressing 2D singularities and achieving better Peak Signal to Noise Ratio (PSNR), but very little attention has been given to the multi-scale and multi-directional attributes of the images and also to further representing the edges and the other anisotropic objects which are the dominant features in medical images. Based on this motivation derived from the current scenario in the field of medical image compression, this research work is focused to deliver a better image compression model for medical images taking the lacuna in the extant methods into consideration. The transform-based image compression models mainly have three common components including transformation tool to decompose the source image, a quantizer to map input values to a finite set of representation and finally the encoder component to encode the image. Three hybrid transform based image compression models are proposed in this work to attain better compression ratio without losing the essential features of the source images. In the first proposed work, a variant of shearlet transform called Cone-Adapted Discrete Shearlet Transforms (CAST) and RIPPLET Type II, a variant of Ripplet transform is combined together with the SPIHT encoder [28] to compress the medical image newline