A low-rank sparse optimization target segmentation method under Laplacian regularization constraint

Abstract

The invention provides a low-rank sparse optimization target segmentation method under Laplacian regularization constraint. In a first frame of a video image, In the case of a known target contour tobe segmented, extracting the contour of a set target in a subsequent video frame on line by using a video segmentation technology; Firstly, segmenting each frame of image; extracting super pixels andcalculating hierarchical convolution characteristics of each super pixel; Establishing a feature matrix of an image, then, known information and segmented targets are utilized; establishing or updating a feature template, solving the optimal expression mode of the current image to the template by using a low-rank sparse optimization algorithm under the Laplace regularization constraint, establishing a saliency map according to the solved expression coefficient, and finally performing accurate segmentation on the contour of the established target by using an energy minimization principle. The method has the characteristics of low calculation complexity and high segmentation precision, and is particularly suitable for the field of single-target online segmentation in video images. The methodhas very high popularization and application values.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to an object segmentation method for low-rank sparse optimization under Laplace regular constraints. Background technique [0002] Video target segmentation technology is to extract the outline of the given target in the video. It is widely used in behavior recognition, motion estimation, target recognition and tracking systems as an image preprocessing process. rapidly developing field. The segmentation of the target in the video is usually considered as a binary classification problem, where the target to be segmented represents the foreground, usually represented by 1, and the other parts represent the background, usually represented by 0. The key to video object segmentation is temporal and spatial consistency. Temporal consistency describes the similarity of objects in consecutive frames, and spatial consistency describes the ability to distinguish objects and backgr...

AI Technical Summary

Problems solved by technology

[0003] At present, many algorithms at home and abroad adopt a step-by-step method in calculating the space-time consistency, and the step-by-step method of this method will lead to the calculation of subsequent steps due to the calculation error of the previous step. For example, the step-by-step method in the prior art Steps: First, establish a template in the feature space for the target to be tracked or segmented according to the known conditions, and then find the area in the feature space where the subsequent image is similar to the template, and output this area as the result. Both methods use In the calculation process, only the similarity between each element feature and the template is considered, and the spatial correlation between the calculated elements is not considered, which makes the target segmentation accuracy not high, especially in the target boundary area. , so there is an urgent need for a new video object segmentation method to further improve it

Images