Human body foreground segmentation algorithm combined with neural network and edge detection

Abstract

The invention discloses a human body foreground segmentation algorithm combined with neural network and edge detection. A Deeplabv3 + algorithm is used to pre-segment human a body image, by using multi-direction detection operator to improve canny edge detection algorithm, an edge correction channel based on Deeplabv3 + segmentation and canny edge detection is proposed, the canny algorithm solvesthe problem that the gradient can only be calculated by the horizontal vertical direction, and the calculation is not accurate because the canny algorithm only calculates the gradient once. The algorithm can provide more directional calculation, and the left upper and right lower directions are combined with the horizontal vertical direction to improve the accuracy of the gradient calculation. This algorithm adopts the combination of neural network Deeplabv3 + and traditional algorithm canny edge detection to segment the foreground of image. The part which can be missegmented can be effectively detected; this algorithm adopts an edge correction channel, deletes the error segmentation accurately, and improves the accuracy of the algorithm.

Description

technical field [0001] The invention relates to the field of human body foreground segmentation algorithm, in particular to a human body foreground segmentation algorithm combined with neural network and edge detection. Background technique [0002] The human body foreground segmentation algorithm refers to the foreground target of effectively segmenting people from images containing human bodies. It is an important branch of image semantic segmentation and plays a basic preprocessing role in computer vision and pattern recognition. It is indispensable for pedestrian authentication and behavior analysis. missing component. With the in-depth study of deep learning in pattern recognition, breakthroughs have been made in human body foreground segmentation. However, due to complex background interference, the problem of edge confusion in semantic segmentation still remains a huge challenge. [0003] Existing human body foreground segmentation algorithms have a problem of high f...

AI Technical Summary

Problems solved by technology

With the in-depth study of deep learning in pattern recognition, breakthroughs have been made in human body foreground segmentation. However, due to complex background interference, there are still huge challenges in the edge confusion of semantic segmentation.

[0003] The existing human body foreground segmentation algorithm has a problem of high false positive rate

At the same time, when simply using a certain foreground segmentation algorithm to process human body images under complex backgrounds, it is difficult to distinguish edge pixels, difficult to connect pixels, and the accuracy of human foreground segmentation is greatly reduced.

When encountering boundaries of different regions, especially when foreground characters and backgrounds are handed over, there is often a high false positive rate problem.

Images