Structural Characteristics Matching Non-Local Means Denoising Algorithm

Definitely, the impact that image denoising has had on image processing in general, is undeniable. The existence of noise hinders people from understanding and extracting image information. And the effectiveness of denoising image also affects the following image processing.Non-Local Means algorithm uses the character that similar structure existed in different natural images, by measuring the similarity between the pixels, estimating the true gray value by the weighted average of the pixel values within the search range, then the denoising performance gained fairly improvement. And the key issue is how to measure structural similarity between image blocks. In this thesis an algorithm including two structural characteristics matching NLM denoising is researched and analyzed, and the main contribution to the research are as follows:(1) CV-Kmeans region classification algorithm is researched and analyzed, using Coefficient of Variation to measure the homogeneity of the image, combined with Kmeans clustering algorithm, classify the structural characteristics of the image.(2) The NLM denoising algorithm based on structural characteristics classification is researched and analyzed. Firstly, according to the urgency of structural changes, classify the noisy image. Secondly, measure similarity of different types of regions within the image blocks of different sizes. Finally, the estimated value is achieved by the weighted average of pixels to remove the noise.(3) A structure-adapted block matching NLM denoising algorithm is researched and analyzed. According to defects of the NLM algorithm based on structural characteristics classification, a structure with self-adaptive NLM image denoising algorithm is proposed. Firstly, the introduction of CV-Kmeans classification algorithm is presented, and the image is divided into structural areas and flat areas. In the structural region, based on the average Euclidean distance between the image patch, adapted options is adopted at different scales of patch size. Based on this, a new algorithm for removing image noise is obtained.Experiment result shows that, NLM algorithm based on structural characteristics classification has obvious advantages over classical NLM. Compared with classical NLM, Probabilistic-Patch-based NLM and the NLM denoising algorithm based on structural characteristics classification, the structure-adapted block matching NLM denoising algorithm sharpens its denoising performance, especially in textural image denoising process.