Research Of Image Denoising Based On Wavelet Transformation And NLM With SSIM

Digital image processing is an important field of information technology, and it isunder constantly changing. However, affected by many factors, people use a variety ofmethods for image processing in the process, often subject to random noise interference.Because of the presence of noise, image quality often degraded. At the same time, the noiseis not conducive to feature extraction, and has a great impact to the subsequent imageanalysis. Therefore, denoising has a very important position and value in the field of imagepreprocessing.In this paper, a variety of de-noising algorithm is studied, focusing on the non-localmeans algorithm (Nonlocal Means, NLM), analyzes the advantages and disadvantages ofthe algorithm and its improvements. Secondly, this paper also made some improvement onthe traditional median filtering.In the field of denoising algorithm based on spatial domain, the median filteringalgorithm is a widely adopted nonlinear filtering method. However, when the noise densityis too high, its processing capacity is poor, not preferable to restore the original image.Therefore, we puts forward an improved median filtering algorithm based on iterativemethod of noise detection in chapter three. At first, noise detection iteration is applied tothe noise image, to improve the accuracy of noise detection. Secondly, according to thenoise point which is detected, the size of the filtering window is changed gradually, inorder to restore image with progressive median filtering. Finally, if the image consists oflarger noise cluster, the noise would be replaced with the mean of signals near it. Iterativenoise detection mechanism and gradual filtering strategy of the reformed algorithm can notonly detect the noise but also has a better effect on preserving the non-noise points of theimage information.NLM denoising algorithm, which is the mainly focused object in this paper, introducesa new perspective for image denoising fields. In this paper, the improved algorithm basedon NLM of image denoising has two aspects:(1) The first improvement of NLMmentioned by this paper is on the matching of similar block. The SSIM model is introducedto the NLM comparisons between similar blocks, and an algorithm of NLM-SSIM ismentioned. Because of taking the advantage of SSIM comprehensive characteristics ofsurface features and structure information of image, which can improve the precision of similar collection and generalization ability, the reformed NLM algorithm has betterdenoising effect than the original one;(2) NLM is a good method on spatial domain, butnot in retention of image details. Because that wavelet transformation can separate the flatpart and uneven part, so combine wavelet transformation with NLM to suppress noise. Atthe same time, mathematical morphology can distinguish and eliminate pixel isolated, thususing it to remove noise in the high frequency part is logical. Synthesize those notions, analgorithm of image denoising based on WTM and NLM-SSIM is mentioned. Combinedwith the wavelet transformation and mathematical morphology, the improved NLMdenoising algorithm is used to get rid of the noise in the low frequency part of wavelettransformation; and mathematical morphology is applied to restrain the noise in the waveletcoefficient of the high frequency part. Finally, by means of the inverse wavelettransformation we can obtain the image which retain more image details and less noise aswell.