An Image Denoising Method Based on Geodesic Distance

Abstract

In this paper, an image denoising method based on geodesic distance is proposed. Different from the denoising method in the traditional Euclidean space, this image denoising method measures the similarity between two pixels in the image according to their geodesic distance. The point in the manifold is a Gaussian model constructed by an area in the image. The geodesic distance between the models represents the difference in the average grayscale intensity and in the abundance of details of the two image regions. This makes the difference in the average gray intensity and in the abundance of details of the regions as the similarity between the pixels, and can more accurately measure the similarity between pixels. Experiments show that this method improves the image denoising effect while still keeping the image details well.