Image denoising for reduced-search fractal block coding

This paper examines the process of image denoising to improve the efficiency of the reduced-search fractal block coding (FBC) of greyscale images by reducing the first-order entropy of the image. The reduced-search FBC is a lossy compression technique that exploits the block-wise self-affinity of an image where portions of the image are represented by scaled and isometrically transformed copies of other portions of the image. The efficiency of this process increases with increased redundancy which is the result of lowering the entropy. Image denoising is concerned with separating noise from an image and then suppressing the noise as much as possible without altering the image itself. In this paper spatial smoothing and wavelet denoising are compared. It is shown that denoising increases the efficiency of reduced-search FBC. Spatial smoothing, however, causes a loss of signal that wavelet denoising does not. In either case, the reconstruction qualities of the peak-signal-to-noise ratio at approximately 34 dB and compression ratios of 18.9:1 and higher have been achieved. This is an improvement over the 31 dB and 18.1:1 for non-denoised images