Adaptive Non-local Means Filter Based on Multi-kernel for Complicated Noise

In the paper, we propose a modified denoising filter based on multi-kernel for color images. To compare the similarity of patches, the patch standard deviation is taken to discriminate flat area and edges, which can capture local geometric structures. It gets rid of the effect of highly dissimilar image patches by setting the weights to zero. Then, we add multi-kernel weights to denoising filter. Different kernel parameters are used to remove complicated noise. The experimental results show that the proposed method has superior performance to existing approaches in terms of noise suppression and detail preservation, especially for the case of low-signal-to-noise ratio (SNR). As our future research work, we intend to apply the method to speech and other intelligent recognition system.