Abstract:

With unknown types of image noise, it is difficult to determine the Meanshift smooth window, which leads the details of image to be blurry. To overcome this problem, a multi-scale Meanshift algorithm of image denoising is proposed. This algorithm combines the advantages of 'digital microscope' of Wavelet and the characteristics of Meanshift of non-parametric probability density estimation and rapid template matching. So it is very efficient to remove the unknown noise of a group of actual distance image at night. In the implementation of the algorithm, first, the image is carried out two-dimensional discrete Wavelet transform, and the low frequency sub-image and the detailed high frequency sub-band are decomposed. Then, different from traditional process, high frequency sub-image is kept unchanged, and the smooth algorithm is implemented on the low frequency sub-image. Finally, the noise is removed based on the reconstruction of the decomposed sub-images. The algorithm not only makes up for the defect of the single Meanshift algorithm, which is difficult to determine the smooth window, leading to the image details be filtered, but also solves the denoising problem on a group of actual distance images at night, whose Signal-to-Noise Ratio (SNR) is 34.29. Experiment results show that the proposed algorithm has higher ability to remove noise, and gets a higher SNR.

Key words: computer application, Meanshift algorithm, unknown noise type, image noise removal