An Image Restoration Algorithm with Fuzzy Similarity based on Wavelet Domain Fusion

Abstract

In order to obtain better image restoration effect, take advantages of different algorithms, the paper proposes a novel image restoration algorithm. Different wavelet filters indicate different character of the image, some are suited to smooth region, some are suited to edge, and the others to texture. First, the blurred image is deblurred by different wavelet filters based on two-step iterative shrinkage threshold algorithm, so the multi-restored images can be obtained. Second, for these multi-restored images, using wavelet domain decomposition, the character matrix is computed based on directional window in wavelet domain, and then the fuzzy similarity matrix is obtained by this character matrix. Last, in wavelet domain some pixels of the restored images which have certain similarity are weighted averaged to the fusion image, the other pixels having not certain similarity are directly picked up obeying ‘maximum value selection’ principle. Experimental results show that the improved signal noise ratio of the restored images is averagely 0.2~1dB higher than that using traditional methods. This proves the effect of the proposed algorithm.