一种改进的基于K-SVD字典的图像去噪算法

An improved image denoising algorithm based on K-SVD dictionary

为了更好地实现图像的去噪效果,提出了一种改进的基于K-SVD(Singular Value Decomposition)字典学习的图像去噪算法。首先,将输入的含噪信号进行K均值聚类分解,将得到的图像块进行稀疏贝叶斯学习和噪声的更新,当迭代到一定次数时继续使用正交匹配追踪(Orthogonal Matching Pursuit,OMP)算法对图像块进行稀疏编码,然后在完成稀疏编码的基础上通过奇异值分解来逐列更新字典,反复迭代至得到过完备字典以实现稀疏表示,最后对处理过的图像进行重构,得到去噪后的图像。实验结果表明,本文的改进算法相对于传统的K-SVD字典的图像去噪能够在保留图像边缘和细节信息的同时,更有效地去除图像中的噪声,具有更好的视觉效果。

In order to achieve the effects of image denoising better, the improved image denoising algorithm based on K-SVD dictionary is designed in this paper. First, the input noisy signal is decomposew3d by using K-means clustering. With the decomposing image blocks, the signal will have sparse Bayesian learning and noise updating. When iterating the given numbers, the signal continues to use OMP algorithm in order to realize the sparse coding. With the completion of sparse coding, the method will update the dictionary by columns using singular value decomposition, iteratively to achieve the completely dictionary and finish sparse representation of image. Eventually, the method will restore original image and obtain the denoising image. Different kinds of images with different noise levels are used to test the algorithm. The experiments and results show that, comparing to the traditional K-SVD dictionary image denoising , our method has better ability of denoising under the premise of keeping the information of image edge and detail and has better vision effection.