Image compression with denoised reduced-search fractal block coding
The research reported in this thesis improves the efficiency of the reduced-search fractal block coding algorithm of greyscale images. The problem addressed here is that additive noise increases the first-order entropy of the image. The increased entropy is equivalent to a lower level of redundancy in the image, and therefore a lower efficiency of the reduced-search fractal block algorithm. This thesis examines a technique of reducing the first-order entropy with a minimum distortion of the signal itself. This is in contrast to spatial filtering techniques such as smoothing which affect not only the noise but also the signal. The technique used in this thesis is called wavelet denoising. Wavelet denoising involves performing the forward discrete wavelet transform of the image, reducing the coefficients by some specified threshold, and performing the inverse discrete wavelet transform. The resulting image, with a lower entropy, is then coded by the reduced-search fractal block compression algorithm. This approach increases the compression ratio by 9.1% to 11% with an acceptable image reconstruction quality.