Edge preserving multiresolution MRF (Markov Random Field) model image segmentation method

Abstract

The invention discloses an edge preserving multiresolution MRF (Markov Random Field) model image segmentation method, which comprises the following steps that: firstly, on the basis of a multi-grid model, establishing a locally interacted image multiresolution grid segmentation model; then, utilizing a Cauchy model with an edge preserving function to extract the edge priori knowledge of the image;and establishing a local area interaction multiresolution MRF model which is combined with edge preserving to segment the image. Therefore, the fusion of the local area characteristics and the edge characteristics of the image is realized, the blocking effect phenomenon of the conventional multiresolution MRF model in an optimization process is solved, and the edge of an image segmentation resultis effectively kept.

Description

technical field [0001] The invention belongs to the technical field of image segmentation, and in particular relates to an image segmentation method of a multi-scale MRF model with an edge preserving function. Background technique [0002] Image processing methods based on multi-scale MRF (Markov Random Field) models have been widely used. This multi-scale MRF structure often adopts the multi-resolution method of the image, using the lower-resolution image to describe the global features of the image, and the higher-resolution image to describe the detailed features of the image, and then through the causality between the layers of the multi-scale MRF model Relationship, build a top-down image segmentation algorithm. [0003] The image segmentation algorithm based on the multi-scale MRF model has low computational complexity, but the quadtree structure of the commonly used multi-scale MRF model leads to blurring or even loss of edge details in low-resolution images during t...

AI Technical Summary

Problems solved by technology

[0003] The image segmentation algorithm based on the multi-scale MRF model has low computational complexity, but the quadtree structure of the commonly used multi-scale MRF model leads to blurring or even loss of edge details in low-resolution images during the establishment of the multi-scale model. , and in the inference process of the multi-scale MRF model, the label information transfer between layers often adopts the direct expansion mapping method

This direct mapping method often leads to blurred edges in the image segmentation results.

[0004] Felzenszwalb (Felzenszwalb P.F, Huttenlocher D P. Efficient Belief Propagation for Early Vision [J]. International Journal of Computer Vision, 2006, 70(1): 167-181.) proposed a multi-scale technology that does not change the image resolution, In the original image, this technology performs multi-scale random field modeling on the image through the multi-grid method of different scales, thus effectively maintaining the detailed features of the image in the larger-scale image, but this method is difficult in the image segmentation process. In , due to the smoothing effect of the prior model of the local area, it will still cause the block effect of the local area and the blurring of the edge of the image segmentation

Images