A Multi-object Labeling Method for Image Scenes Based on Prior Condition Constraints

Abstract

The invention discloses a multi-object labeling method for an image scene based on a priori condition constraints, including: determining the region of interest of a semantic object group; calculating the multi-dimensional features of a test image, as a priori appearance constraints, and transforming the pixel-level multi-dimensional features It is a superpixel-level multi-dimensional feature; construct a graph model structure of the region of interest in the test image, use the superpixels in the region of interest as the graph structure nodes, use the adjacency relationship of the superpixels as the edges of the graph structure, and use the corresponding features of the prior appearance constraints Convert to edge weight value, calculate the initial geodesic distance, as the node weight value; perform geodesic propagation, determine the object label of the current seed point in each step of propagation, and update the geodesic distance of adjacent points around it, as The next step is to prepare for the propagation until the end of the propagation process to obtain the object label of each superpixel. By adopting the technical solution of the present invention, the rich features of objects are used as prior constraints to improve the accuracy of object marking.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a multi-object labeling method for image scenes based on prior conditional constraints. Background technique [0002] With the rapid development of social science and technology, electronic devices such as smartphones, tablet computers, and cameras are increasingly used in social life. Accompanying it is that the acquisition of image data is becoming more and more convenient, and the amount of data is also increasing. People The demand for image processing and applications is also becoming more and more abundant, and various image processing software tools are also emerging. In various industries that promote economic and social development, the demand for image scene understanding has received more and more attention. For example, in unmanned driving systems, it is necessary to understand street scenes, identify lane lines, traffic signs, and obstacle detection. , to g...

AI Technical Summary

Problems solved by technology

[0005] However, it should be noted that the semantic segmentation of objects is very difficult. It not only requires the basis of semantic recognition to distinguish between categories, but also requires the division between multiple objects within the category, and multiple objects often show various differences

At present, the work in this direction is still in an exploratory stage, and further in-depth research on theory and technology is urgently needed

Images