Research On Multi-channel Color Image Denoising Algorithms

Image denoising is one of the hotspots in digital image processing tasks, image denoising effective algorithm is the key of digital image information pretreatment. Effective denoising algorithm can not only improve the image edge detection, image match, image fusion, but also is conducive to the application of the aerospace, information engineering,biological medicine, culture art and robot vision, etc. With the application of image denoising, the multi-channel color image denoising becomes a new requirement. This thesis based on the national natural science foundation of China (61379010), research on multi-channel color image denoising algorithm in frequency domain.The main work is as follows:1. Being frequency domain image denoising algorithm,there is a problem that when larger noise variance exist in the image, the denoising effect is not good. An improved Normal Inverse Gaussian (NIG) distribution model of image denoising algorithm is proposed, the algorithm use optimal threshold linear interpolation function to improve NIG as the coefficient of priori distribution model. Simulation experiments show that the proposed model can be used to accurately model the coefficient of heavy tailed image with large noise standard deviation and improve the denoising effect.2. For limitations of the correlation between channel of traditional color space, the chromatic aberration exist in color image after denoising. A method of reducing the channel correlation is proposed. The method uses principal component analysis to reduce channel correlation based on frequency domain decomposition coefficients, getting three directions of vector as an adaptive color space. The experimental results show that the correlation between the constructing color space channel is lower, so this color space can improve denoising effect.3. After using color image denoising algorithm, denoised image still has residual noise when the noise variance is higher. In order to solve this problem, based on the adaptive color space, using frequency domain performance good single channel denoising algorithm can further enhance color image denoising effect. The experimental results show that this method can not only remove higher standard deviation noise, but also the denoised image still clear, it can reflect effectiveness of the method.