一种基于差异系数的稀疏度自适应图像去噪算法

基于压缩感知的K-means Singular Value Decomposition(K-SVD)图像去噪算法具有良好的自适应性和细节恢复能力,但需事先给定稀疏度K。 该方法的去噪效果会受到图像稀疏度的影响。另外,训练初始系数时用到的追踪类算法中通过向量内积值的大小评定图像分量间相关度的方法,因存在大值噪声点,容易造成假相关,从而影响去噪效果。提出基于差异系数的稀疏度自适应K-SVD去噪算法,通过引入差异系数来平衡因噪声点造成的假相关问题,同时使用相关度均值作为阈值来自适应地产生稀疏度K,避免因给定不恰当的稀疏度而影响去噪效果的问题。在USC标准库上的实验结果表明,所提算法在去噪效果方面有一定的优越性。

Sparsity-adaptive Image Denoising Algorithm Based on Difference Coefficient

With the remarkable adaptability and the details recovery capability,K-SVD is a highly effective method based on sparse representation theory in image denoising.But the sparsity K should be given in advance,and different images have different K values in fact.On the other hand,pursuit algorithms which are used in evaluating the relevance between vectors of an image by calculating vector inner product,are brought into K-SVD to train sparse coefficients.Denoising effect is reduced because a few noisy pixels are likely to cause false relevance.This paper addressed this problem and proposed a novel sparsity-adaptive speeded K-SVD(SASK-SVD) algorithm based on different coefficient,which can improve the efficiency.The different coefficient is to eliminate false relevance.The sparsity K is adaptively generated by using the average correlation as the threshold.This study conducted extensive experiments to demonstrate these ideas.The experimental results show that the proposed method achieves the state-of-the-art denoising performance.