Wavelet-based image enhancement
This thesis proposes the novel wavelet-based digital image denoising methods to solve the problems which are difficult for the best known two-dimension adaptive Wiener filtering technique. Because the two-dimension Wiener filtering requires a lot of computations, it becomes unsuitable for real-time processing environments. Our algorithm solves this problem by first transforming the image into wavelet transform domain subimages, then we process the reduce-sized subimage with fewer operations for one pixel and fewer total pixels to be processed. After processing the subimage, we transform it back to the spatial domain. Experiments show that our algorithm is much faster at processing the images degraded by Additive Gaussian White Noise (AGWN) than the two-dimension adaptive Wiener filtering technique while achieving the same processing qualities. Further, the Wiener filtering technique is a frequency selective attenuation method. It performs a certain amount of attenuation at fixed SNR( f ) frequency according to the ratio (SNR(f))/(SNR(f) + 1) where SNR(f) is the signal to noise ration at that frequency. So, for images degraded by AGWN with extremely high power, the Wiener filtering technique will perform severe attenuation at each frequency. Thus, the two-dimension Wiener filtering cannot achieve satisfactory result in this case. Our novel algorithm solves this difficulty by two-level wavelet transforming the noisy image first, then we transform the wavelet coefficients to the range of 0-255 gray levels and treat the resulting subimage as an ordinary spatial domain image. Based on that spatial domain image, several modified conventional spatial domain denoising techniques are applied to it. Experiments show that our algorithm can improve the image quality a lot while the two-dimension adaptive Wiener filtering cannot.