基于自适应阈值的三维块匹配降噪算法研究

发布时间：2019-02-09 16:34

【摘要】：图像在人们生活中的影响越来越大,是人们相互交流的重要手段。然而在图像的生成、传输和获取的过程中,由于外界和内在因素的干扰,图像常常被加入了噪声,影响了人们对图像识别的视觉效果。所以,图像降噪是图像处理中十分重要的部分。图像降噪的首要目的是保留图像的结构信息,然后才是对图像进行降噪。无论是空间域降噪算法还是频率域降噪算法,在去噪时往往忽略了图像的边缘细节,导致降噪后的图像变得模糊。三维图像邻域块匹配降噪方法(BM3D)是目前最出色的算法之一,它解决了图像在降噪后模糊的问题。然而由于其采用硬阈值进行图像相似邻域块的匹配,在图像噪声强度较高的情况下,阈值相对较低,导致图像相似块获取数量不足,图像降噪效果不理想;在图像噪声强度较低的情况下,阈值相对较高,图像相似块匹配数量过多,图像降噪时间过长。本论文在深入研究图像降噪的基础和理论后,针对BM3D降噪的缺点,提出了一种改进算法——基于自适应阈值的BM3D降噪方法。该算法的核心是在图像相似邻域块的获取过程中,使用自适应的阈值代替原BM3D方法中的距离硬阈值。对原始图像进行邻域分块,对选出的参考邻域块计算与其他图像块的梯度结构相似以及欧式距离,将这两者进行对比,匹配时的阈值就是图像块最相似时的欧式距离。该算法降噪后的图像无论是客观测评还是主观视觉观察,都取得了比较好的降噪结果。
[Abstract]:Image is becoming more and more important in people's life, and it is an important means for people to communicate with each other. However, in the process of image generation, transmission and acquisition, the noise is often added to the image due to the interference of external and internal factors, which affects the visual effect of image recognition. Therefore, image denoising is an important part of image processing. The primary purpose of image denoising is to preserve the structure information of the image, and then to reduce the image noise. In both spatial and frequency domain denoising algorithms, the edge details of images are often ignored when denoising, resulting in the blurred image after denoising. Neighborhood block matching denoising (BM3D) is one of the most excellent algorithms in 3D image, which solves the problem of image blur after denoising. However, because the hard threshold is used to match the image similar neighborhood block, the threshold value is relatively low when the image noise intensity is high, which leads to the lack of the number of image similar blocks, and the image denoising effect is not ideal. In the case of low image noise intensity, the threshold is relatively high, the number of image similar blocks matching is too much, and the image denoising time is too long. In this paper, after deeply studying the foundation and theory of image denoising, an improved BM3D denoising method based on adaptive threshold is proposed to overcome the disadvantages of BM3D denoising. The core of the algorithm is to replace the distance hard threshold in the original BM3D method with adaptive threshold in the process of obtaining the image similar neighborhood block. The original image is divided into neighborhood blocks and the selected reference neighborhood blocks are calculated similar to the gradient structure and Euclidean distance of other image blocks. The threshold of matching is the Euclidean distance when the image blocks are most similar. After denoising, both objective evaluation and subjective visual observation have been achieved.
【学位授予单位】：中北大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】