稀疏表示去噪算法研究

发布时间：2018-05-19 12:39

本文选题：稀疏表示 + 形态学成分分析　；参考：《河北工业大学》2014年硕士论文

【摘要】：信号稀疏表示是一种新兴的信号分析和综合方法，其目的就是在过完备字典中用尽可能少的原子来表示信号。稀疏表示的应用广泛并已成功应用于图像去噪领域，在稀疏表示的过程中即可将噪声滤除。稀疏表示去噪的方法可以分为基于局部和非局部的方法。K-SVD是局部的去噪方法，，NLM、BM3D是非局部的去噪方法。本文针对局部和非局部的去噪方法介绍了两种稀疏表示的方法来对图像去噪。针对局部算法K-SVD在去噪的时候对纹理性强的图片效果好，对于结构性强的图片处理效果不佳的问题，提出一种新的去噪方法，即形态学成分分析去噪算法，将图像分为结构层和纹理层。对纹理层采用K-SVD去噪，对于结构层采用BM3D的去噪，将二者的结果合并作为图像的最后去噪效果。实验结果表明该算法对图像去噪的效果要好于单一使用K-SVD和BM3D去噪的效果。针对不相邻但结构相似图像块存在冗余性的问题，提出一种聚类的方法，将聚类加入到非局部去噪的方法中。建立新的关于聚类的模型，在求解模型中不断地更新聚类中心和PCA字典。实验结果表明该算法有效的利用了图像块间的冗余，图像去噪的效果要好于K-SVD的局部去噪效果。
[Abstract]:Signal sparse representation is a new signal analysis and synthesis method, which aims to represent signals with as few atoms as possible in overcomplete dictionaries. Sparse representation is widely used and has been successfully applied in the field of image denoising. In the process of sparse representation, the noise can be removed. The sparse representation denoising method can be divided into local and non-local denoising methods. K-SVD is a local denoising method. In this paper, two sparse representation methods for image denoising are introduced for local and non-local denoising methods. A new denoising method named morphological component analysis (MCAA) is proposed to solve the problem that the local algorithm K-SVD has good effect on textured images and poor structural image processing. The image is divided into structure layer and texture layer. The texture layer is de-noised by K-SVD, and the structure layer is de-noised by BM3D. The results of the two methods are combined as the final denoising effect of the image. Experimental results show that the algorithm is better than K-SVD and BM3D in image denoising. Aiming at the redundancy of non-adjacent but structurally similar image blocks, a clustering method is proposed, which is added to the non-local de-noising method. A new clustering model is established and the clustering centers and PCA dictionaries are constantly updated in the solution model. The experimental results show that the algorithm makes use of the redundancy between image blocks effectively and the effect of image denoising is better than that of K-SVD.
【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.73

【参考文献】