Improving the Quality and Contrast of Image Details Using the Geodesic Distance Filter

Abstract —As a rule, modern methods of improving and contrasting image details are based on the edge-preserving filters or bilateral filters. However, the computational complexity of classic bilateral filters is proportional to the square of the number of pixels in the image and fast algorithms are not always efficient enough or the filtering result does not always match the result of the original filter. We propose replacing the classic bilateral filter with a geodesic distance filter, which also belongs to the class of convolution transformations that can improve visual perception of images using information about the edges of objects in the processed image. The convolution kernel based on the geodesic distance has several advantages, as it allows recursive computation and, therefore, fast image processing. We also propose a method for suppressing color artifacts by switching from the standard RGB representation to the HSV representation, which is more balanced with respect to the perception of human vision. The effectiveness of the proposed filter is compared using the illustrations to this article so that the reader can visually compare the qualities of various processing options.