Image segmentation level selection method based on scale perception

Abstract

The invention discloses an image segmentation level selection method based on scale perception, and aims to improve the segmentation quality of an overall image. The method comprises the following main steps: firstly, obtaining a segmentation result represented by a tree form by using a multi-level image segmentation algorithm; secondly, calculating feature vectors of all segmented regions in eachsegmentation layer, and performing quantitative description on region segmentation quality according to a visual format tower principle; thirdly, constructing a graph model of the multi-level segmented image by taking the hierarchical region with the finest segmentation granularity as a node, and finally, mapping a label to a region corresponding to the original hierarchy, and combining to obtaina final image segmentation result. According to the method, based on the scale selection principle of segmentation area quality, the problem of hierarchical selection of multi-level image segmentation is solved, and the limitation of traditional single hierarchical selection and the uncertainty of threshold parameters on segmentation hierarchical selection are overcome. As a post-processing means, the output quality of the multi-level image segmentation algorithm in a visual processing task can be improved.

Description

technical field [0001] The invention relates to the technical fields of computer vision and image processing, especially multi-level image segmentation technology. Background technique [0002] Image segmentation is a key issue in image processing and computer vision. It usually refers to dividing an image into several disjoint areas according to certain standards, so that the grayscale, color, texture and other features of the image in the same area show consistency or similarity. However, these characteristics show obvious differences among different regions. Image segmentation embodies the understanding of image content, but due to the lack of uniform standards, meaningful objects in images often have multi-level (scale) characteristics. Therefore, image objects can be extracted with different granularities and semantics, and different numbers of regions can be used to describe the same object. Based on this idea, people proposed a multi-level image segmentation method,...

AI Technical Summary

Problems solved by technology

There are many problems in this method: firstly, the image may contain multiple targets, and their optimal scales may appear in different segmentation levels, so it is impossible to use a single threshold to find all the optimal target areas; secondly, the threshold selection depends on expert design. It is not only cumbersome but also subject to subjective differences, or relying on existing labeled data, it is impossible to achieve online image segmentation

If the level is selected by the overall segmentation, it is equivalent to the expert threshold setting, but it does not necessarily meet the optimal scale of each target area.

Images