Video object cooperative segmentation method based on track directed graph

Abstract

The invention relates to a video object cooperative segmentation method based on a track directed graph. The method comprises the following steps of (1), inputting each frame sequence Fm, t(t=1, ..., Nm) of a video set; (2), generating a motion vector field, an initial significance graph and a candidate object for the video frame Fm, t; (3), performing frontward and backward tracking on each candidate object, and performing maximal inhibition and track segmentation, thereby forming a track set; (4), constructing a directional weighted graph G=(V,E), wherein the track is a directional edge which is established between nodes in the graph according to a matching score; (5), converting the directional weighted graph G=(V,E) to a non-directional weighted graph G=(V,E'), extracting a maximal clique by means of a maximal clique extracting algorithm, calculating the weighted clique score of each maximal clique, using a track area which corresponds with the clique with highest weighted clique score as a main object area, performing a popular sequencing algorithm for generating an objective significance graph, and obtaining a final segmentation result by means of GrabCut; and (6), according to the obtained object segmentation result, updating an initial significance graph, calculating the maximal clique score and obtaining the object of other kind.

Description

technical field [0001] The invention relates to the technical field of image and video processing, in particular to a method for collaborative segmentation of video objects based on a trajectory directed graph. Background technique [0002] With the vigorous development of the Internet and multimedia technology, the means for people to obtain images is becoming more and more convenient and flexible. For example, in surveillance video, social network, news reports and other fields, video data is gradually increasing, and video data contains richer information. In the process of data explosion, people's demand for intelligent processing of video data is also increasing. Video object segmentation is a challenging field, which aims to extract the main object area that people pay attention to in the video. Compared with image segmentation, video object segmentation uses the motion information of the object in the video, which can extract meaningful information more effectively. ...

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is to propose a method for collaborative segmentation of video objects based on trajectory directed graphs for the defects in the prior art. This method can be used without manually setting the similarity threshold of common objects between videos. More accurate and automatic extraction of multi-type objects appearing in video groups

Images