Efficient Video Object Segmentation based on Local-Global Consistency Learning

Efficient video segmentation algorithms are urgently needed as the rapid development of the multimedia technology has made real-time video data the mainstream of information transmission and analysis. This research work proposes a novel video object segmentation based on local-global consistency learning. In the designed model, local-global consistency learning algorithm is proposed. The core idea is to pass on the label information of each sample point to its neighboring samples through an iterative algorithm until all of the samples have reached a steady state of the label. Then, by applying this into the video frame analysis, the efficient segmentation is achieved. To solve the problem of information integration, the mixed features of the coded output are mapped to the frequency domain for the decomposition. By using the average precision and recall, the test is conducted. The performance is validated and the visualized results on the different collected images are tested.