聚类算法在图像修复上的应用

发布时间：2018-09-04 05:57

【摘要】：图像修复是图像处理领域中一项重要技术,主要是利用图像中现有的信息,估计图像中的未知区域,从而使修复后的图像在视觉上与原图像接近,图像修复技术在实际生活中应用广泛。基于稀疏表示的图像修复技术已经成为一个新的研究方向,本文是在稀疏表示理论研究基础上,结合字典训练和聚类方法对图像修复技术进一步研究。本文首先介绍了图像修复的研究背景、意义以及国内外研究现状,主要研究内容是:介绍稀疏表示理论、常用的稀疏表示算法和K-SVD字典学习算法;分析了目前常用的图像修复算法的优缺点。详细地分析了基于稀疏表示和字典学习的图像修复算法,采用此方法进行图像修复,其结果的优劣性与学习的字典的好坏有很大关系,本文为了保证字典内容丰富且具有自适应性,在进行字典学习前,先对图像库的图像进行聚类,因此本文还介绍了模糊聚类算法及其改进算法,使用Canopy算法改善模糊聚类算法以及在分布式平台上的应用。仿真结果表明Canopy-FCM算法相比FCM算法准确率提高,簇间距离减小,加入聚类后的图像修复算法在均方根误差和峰值信噪比的数值上,优于仅仅利用待修复图像本身进行字典训练的图像修复算法,算法很好地利用了图像之间的相似性,对含有丰富结构信息的图像,修复效果较好,并且弥补了基于K-SVD的图像修复算法的一些缺憾,在一定程度上改善了图像修复的质量。
[Abstract]:Image restoration is an important technology in the field of image processing. It mainly uses the existing information in the image to estimate the unknown region in the image, so that the restored image is close to the original image visually. Image restoration technology is widely used in real life. Image restoration based on sparse representation has become a new research direction. Based on the theory of sparse representation, this paper combines dictionary training and clustering method to further study image restoration technology. This paper first introduces the research background, significance and current situation of image restoration. The main research contents are: sparse representation theory, sparse representation algorithm and K-SVD dictionary learning algorithm; The advantages and disadvantages of current image restoration algorithms are analyzed. The image restoration algorithm based on sparse representation and dictionary learning is analyzed in detail. In order to ensure that the dictionary is rich in content and self-adaptive, the image of the image library is clustered before the dictionary is studied, so this paper also introduces the fuzzy clustering algorithm and its improved algorithm. Canopy algorithm is used to improve fuzzy clustering algorithm and its application on distributed platform. The simulation results show that the accuracy of Canopy-FCM algorithm is higher than that of FCM algorithm, and the distance between clusters is reduced. The RMS error and peak signal-to-noise ratio (PSNR) of the image restoration algorithm after clustering are calculated. The algorithm is better than the image restoration algorithm which only uses the image itself to be repaired for dictionary training. The algorithm makes good use of the similarity between the images, and has a good effect on the images with rich structural information. It also makes up for some shortcomings of image restoration algorithm based on K-SVD, and improves the quality of image restoration to some extent.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP311.13

【参考文献】