改进的K-奇异值分解图像去噪算法

Improved K-SVD Image Denoising Algorithm

针对传统的K-奇异值分解信号利用率不足,采用了稀疏贝叶斯学习预处理图像信号;将正交匹配追踪与改进之后的最速下降理论相结合;因噪声原子存在于字典更新之后得到的字典中,所以结合Bartlett检验法将噪声原子裁剪掉。实验结果表明,此方法相对于小波阈值去噪法、基于离散余弦变换字典稀疏表示等去噪方法能够更好地滤除噪声,保留图像边缘信息,获得更高的峰值信噪比,得到图像视觉效果更佳。

Aiming at the problem of shortage of signal utilization by traditional K-SVD. Using sparse Bayesian learning to preprocess the image signal. Combining the orthogonal matching pursuit algorithm with the improved steepest descent algorithm. Taking into account the noise atoms are present in the dictionary after the update of the dictionary, so combined with the Bartlett test method to cut off the noise atoms. Experimental results show that the method is better than the wavelet threshold denoising algorithm and the sparse representation based on DCT dictionary. Also, the method can remove the noise better, preserve image edge information, obtain higher peak signal to noise ratio, and the resulting image has a better visual effect.