The Research On Image Denoising Algorithm Based On Non-local Mean

During the acquisition and transmission of the digital images, the images are often influenced by noises. Therefore how to effectively remove the noise is a hot field in the image processing. So far, lots of researchers have proposed various denoising methods. This thesis does lots of research on several denoising methods, and focus on Non-Local Mean. We analysis its strengths and make some improvements on weaknesses. The main research contents are outlined as follows:Firstly, an improved kernel weights of non-local means denoising algorithm is proposed. The Gaussian-Turkey function is adopted as a weight function of non-local means algorithm. Compared with the exponential function, It has a faster decay rate, that is to say, when the distance increases, it fast decay, even the weight is set to zero. Experimental results show that the image PSNR is obviously improved。Secondly, we propose an adaptive-fast non-local means denoising algorithm. Using Joint edge detection means detected edge points and non-edge point of the image. the edge points use wavelet threshold de-noising and others use non-local mean. Thus we get the final estimate image signal. Experimental results show that the new algorithm significantly improved the speed.Finally, we propose a two-stage non-local means feedback algorithms. This can extact image detail from methods noise as much as possible. Experiments show that while removing the noise, the algorithms can maximize the retention of the image detail.