【摘要】 图像复原是图像处理中一个重要的研究课题。大部分图像复原算法,都只是利用图像单个像素点,或某一邻域内的灰度和梯度信息。如果在图像复原模型中能够超越邻域,而更大范围地利用图像内容本身的信息,将可能更有效地改善复原质量。实际上,大多数自然图像中,其内容具有自相似特性。基于非局部的图像处理方法,就是以这类自相似特性为出发点的。提出一种改进的基于非局部的正则化图像去噪算法,主要针对非局部处理方法中的权值计算作了改进处理。与原始算法不同,采用EMD来进行图像子块间的相似度计算,并结合图像的方向信息分类子块区域,以解决计算量过大的问题。更多还原

【Abstract】 Image restoration is an important research subject in image processing.Image gray and gradient magnitude based on single pixel or a neighbour is used in most of image restoration algorithm.If these information,which came from more wide-ranging image region,can be used for image restoration model,it is possible to improving restoration quality.In reality,there are self-similarity about image content in natural image.It is a foundation for non-local image processing method.This paper presented an improved image denosing algorithm based on non-local regularization.Designed the improvement for computing weight which camed from non-local regularization.Used the earth mover’s distance to compute the similarity between image sub-region.At the same time,used to solving expensive compute,the image direction information to classify image sub-region.更多还原