Abstract

With the advent of digital technology, the amount of digital medical images generated is increasing at an exponential rate. When used in interactive telemedicine applications, these raw images need a huge storage unit, thereby necessitating the use of highbandwidth medium for transmission. In order to reduce the cost of transmission and enable archiving, especially for medical applications, image compression is performed. Most of the existing research in compression is focused on minimizing mathematically tractable, easy to measure, distortion metrics like the peak signal-to-noise ratio (PSNR) for gray scale and root mean square error (RMSE) for color coded images. These metrics often poorly correlate with the visual quality of the image as the human visual sensitivity characteristics are not incorporated in these evaluation techniques. For digital images, especially for medical images, perceptual quality of image is important along with compression efficiency since diagnosis is done based on the visual attributes of the reconstructed image. This thesis is concerned with the problem associated with obtaining an enhancement in the quality of the image with lossy compression schemes. An enhanced image compression scheme using the nearly-orthogonal wavelet to achieve an enhancement in the perceptual quality of the image is proposed. This approach differs from the conventional design scheme in that it incorporates the human visual system (HVS) characteristics directly in each sub-band of the wavelet decomposed image. The enhancement in the visual quality of the reconstructed image is achieved by using the newly derived contrast sensitivity function band peak-average (CSF-BPA) mask at each decomposition level during wavelet transformation phase in the compression system. The nearly-orthogonal wavelet is used to achieve improved compression performance for low and medium bit rate applications. The proposed scheme is further extended for perceptual medical image compression application for images such as x-ray, MRI and ultrasound scan. The dependency of the calculated PSNR to the observed visual quality has been addressed for some of the standard images as well as for medical images. The presence of additive white Gaussian noise (AWGN), which is the most common source for image artifacts, is considered during the evaluation of the performance of the scheme. The effectiveness of the proposed scheme is demonstrated and compared with that of the conventional scheme through extensive quantitative and qualitative assessments. Finally, a multiplier-less design for the implementation of the BNC 17/11 wavelet filter in a FPGA or ASIC hardware is proposed