Adaptive Fusion Based Hybrid Denoising Method for Texture Images

This paper presents an efficient image denoising method by adaptively combining the features of wavelets and wave atom transforms. These transforms will be applied separately on the smooth areas of the image and the texture part of the image. The disintegration of the homogenous and nonhomogenous regions of noisy image is done by decomposing the noisy image into a noisy cartoon (smooth) image and a noisy texture image. Wavelets are good at denoising the smooth regions in an image and will be used to denoise the noisy cartoon image. Wave atoms better preserve the texture in an image hence is used to denoise the noisy texture image. The two images will be fused adaptively. For adaptive fusion different weights will be chosen for different areas in the image. Areas containing higher degree of texture will be allotted more weight, while the smoother regions will be weighed lightly. The information regarding the weights selection will be obtained from the variance map of the denoised texture image. Experimental results on standard test images provide better denoising results in terms of PSNR, SSIM, FOM and UQI. Texture is efficiently preserved and no unpleasant artifacts are observed.