非局部均值图像去噪算法研究

项目名称： 非局部均值图像去噪算法研究

项目编号： No.61303126

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 胡金蓉

作者单位： 西华大学

项目金额： 22万元

中文摘要： 非局部均值去噪算法基于图像固有的冗余特性、利用图像中含有的大量重复结构、采用空域加权均值滤波的形式来非线性地去噪，具有假设条件弱、表达形式简单、去噪效果好等优点，是当前图像去噪领域内的一个研究热点。本项目拟对非局部均值去噪算法在噪声存在时相似性度量准确度不高和难以自适应地选取算法参数等问题进行深入地研究。主要研究内容包括：⑴研究提高非局部均值算法相似性度量准确度的理论与方法；⑵研究自适应地选取非局部均值算法相似性窗口半径参数和滤波参数的理论与方法；⑶研究自适应地选取非局部均值算法搜索窗口参数的理论与方法；⑷研究将非局部均值算法用于弥散张量磁共振成像数据去噪处理的模型与方法。通过本项目的研究，预期获得的研究成果将提升非局部均值算法的去噪性能、增强算法的实用性并拓展算法的应用范围。

中文关键词： 图像去噪；非局部均值；随机采样；结构张量；磁共振成像

英文摘要： The nonlocal means(NLM) denoising algorithm recovers the clean image from distorted noisy image without any regular assumptions, only based on inherent redundancy of image itself. And it employs abundant of repeated patterns contained in image to nonlinearly denoise by weighted mean filtering in spatial domain. Moreover, the NLM method has merits of weak hypothesis, simple mathematical expression, easy implementation, etc. Currently, it can obttain the best denoising result than other methods demonstrated by theoretical analysis and experiments, and it is a hot research topic in image denoising. In this project, some problems with NLM will be researched profoundly. For example, the calculation of similarity weights for NLM has a limited accurate capability against noise when the noise standard deviation is large, and there is no way to adaptively select values for NLM's parameters. The main research content of this project includes the following four parts: (1) the theories and methods about how to improve the accuracy of NLM's similarity measure will be studied; (2) the theories and methods about how to adaptively assign values to NLM's parameters of similarity window radius and filtering intensity will be studied; (3) the theories and methods about how to adaptively select homogeneous region of current pixel a

英文关键词： image denoising；nonlocal means；random sampling；structure tensor；magnetized resonance imaging