基于BM3D的复杂纹理区域图像去噪

Complex Texture Region Denoising Based on BM3D Image Denoising

纵观图像去噪领域,目前也出现了一些效果较好,并且应用广泛的图像去噪算法,例如频域滤波中的小波分解、遗传算法(GA)、字典学习算法(K-SVD)、非局部滤波(Non-Local Means,NLM)等等,可以说在单图像去噪上效果最被认可的还是块匹配3D去噪(BM3D)。但在该方法应用的过程中,还是出现了一些需要改进的实际问题。特别是在图像复杂纹理区域的去噪上,由于BM3D去噪的结构特性,导致了在该区域上并没有得到较好的效果。所以在该文提出了一种结合TV模型和改进的Prewitt边缘检测的混合去噪方法。最后的实验结果也表明了改进的BM3D方法相对于原始的方法还是取得了不错的提升。

Throughout the field of image denoising, At present, there are some image denoising algorithms with good effects and widely used, such as wavelet decomposition in frequency domain filtering, genetic algorithm(GA), dictionary learning algorithm(K-SVD), and non-local filtering(Non-Local Means, NLM), etc. It can be said that the most recognized effect on single image denoising is block matching 3 D denoising(BM3 D). However, in the process of applying the method, there are still some practical problems that need to be improved. Especially in the denoising of complex texture regions of images, due to the structural characteristics of BM3 D denoising, no good results are obtained in this region. Therefore, a hybrid denoising method combining television(TV) model and improved Prewitt edge detection is proposed in this paper. The final experimental results also show that the improved BM3 D method has achieved a good improvement over the original method.