基于相似块的图像去噪算法研究

发布时间：2018-05-05 09:58

本文选题：相似块 + 非局部均值　；参考：《西北大学》2017年硕士论文

【摘要】：图像噪声是由多种因素造成的一种随机现象。噪声的存在直接影响了图像真实信息的表达,如何从受到噪声污染的图像中恢复原有信息,提升图像质量已经成为图像处理领域的热点研究方向。非局部均值(Non Local Mean,NLM)算法是一种能够有效去除加性高斯噪声的去噪算法,多年来吸引了大量学者的关注。NLM算法将相似块的加权平均推广到图像整个区域,获得了较好的去噪效果。但是,NLM算法在相似块的选择与度量、参数设置和执行效率等方面仍存在一些不足。本文主要围绕NLM算法中相似块的选择和度量开展研究,研究了 NLM图像去噪算法的改进和优化。论文的主要研究工作包括:针对NLM去噪算法在部分区域图像块的相似度较低问题,提出了优化相似块选择策略的改进算法,通过超像素分割算法将图像划分为若干相似块,在变化丰富的区域利用纹理信息直接选择相似度更高的图像块;针对不同区域特点采用了不同尺寸的相似窗口,新的相似块选择策略保留了原图像更多的纹理信息。针对相似度较小的图像块参与像素信息恢复时影响去噪效果的问题,提出了相似块二次筛选和新的权重分配方法。考虑到与中心像素在同一类的图像块间具有更高的相似性,在搜索窗内按相似块的局部特征将其聚类,并对聚类同簇中的相似块赋予更高的权值。由于选择了更加合理的相似块,提升了算法的去噪性能。针对NLM去噪算法中利用像素间的欧氏距离衡量相似度存在的不足,提出了一种新的判定函数用于相似块的度量。通过对相似块多级小波分解,从小波系数中提取能量信息并定义了基于小波的相似度评判函数。仿真实验表明新的相似度度量方法在图像去噪时获得了好的性能,体现了该方法的有效性。仿真实验表明了本文算法的有效性和可行性。
[Abstract]:Image noise is a random phenomenon caused by many factors. The existence of noise has a direct impact on the expression of real image information. How to recover the original information from the image contaminated by noise and improve the image quality has become a hot research direction in the field of image processing. Non-local mean Local (NLM) algorithm is a kind of denoising algorithm which can effectively remove additive Gao Si noise. For many years, it has attracted a lot of scholars' attention. The weighted average of similar blocks is extended to the whole region of the image, and a good denoising effect is obtained. However, the NLM algorithm still has some shortcomings in the selection and measurement of similar blocks, parameter setting and execution efficiency. This paper focuses on the selection and measurement of similar blocks in NLM algorithm, and studies the improvement and optimization of NLM image denoising algorithm. The main research work of this paper is as follows: aiming at the low similarity of NLM denoising algorithm in some regions, an improved algorithm to optimize the selection strategy of similar blocks is proposed, and the image is divided into several similar blocks by super-pixel segmentation algorithm. Texture information is used to directly select the image blocks with higher similarity in the regions with rich changes, and different sizes of similar windows are adopted for the characteristics of different regions. The new similarity block selection strategy retains more texture information of the original image. To solve the problem that image blocks with low similarity affect the denoising effect when pixel information is restored, a new method of weight allocation and quadratic filtering of similar blocks is proposed. Considering that the image blocks of the same class have higher similarity with the central pixels, the similar blocks are clustered according to the local characteristics of the similar blocks in the search window, and the similar blocks in the cluster are given higher weights. Because more reasonable similar blocks are selected, the denoising performance of the algorithm is improved. Aiming at the shortcoming of using Euclidean distance between pixels to measure similarity in NLM denoising algorithm, a new decision function is proposed to measure the similarity blocks. The energy information is extracted from the small wave coefficients by multilevel wavelet decomposition of similar blocks and the similarity evaluation function based on wavelet is defined. The simulation results show that the new similarity measure method has good performance in image denoising, which shows the effectiveness of the method. Simulation results show the effectiveness and feasibility of the proposed algorithm.
【学位授予单位】：西北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】