A low-delay video segmentation real-time preview method

Abstract

The invention provides a low-delay video segmentation real-time preview method. The method comprises the following steps: processing a key frame of a video stream by adopting a network structure basedon deep learning to obtain a second image segmentation result; Processing the transition frame between the key frames of the video stream by adopting a gray scale projection algorithm to obtain a second image segmentation result of the transition frame; And displaying the first image segmentation result in real time by adopting a low-delay display strategy. According to the invention, the key frame of the video stream is processed by the deep learning network structure, and image segmentation can be accurately carried out; The gray scale projection algorithm is adopted to process the transition frame between the key frames of the video stream, and the similarity between the video frames can be utilized to rapidly propagate the segmentation result of the previous frame, so that the segmentation time of each frame is short, and the smoothness of the video is ensured; And a low delay strategy combining an accurate result obtained by processing the key frame through the deep learning network structure with the video sequence enables the video to be free of a stuck feeling and a lag feeling.

Description

technical field [0001] The invention relates to the technical field of video segmentation, in particular to a low-delay real-time preview method for video segmentation. Background technique [0002] Image segmentation is an important part of computer vision and has a wide range of applications in real life, such as tissue detection in medical images, disaster assessment, portrait beauty, smart mapping, etc. Video image segmentation refers to the process of segmenting the foreground and background of objects in each frame of video to obtain a binary image. Due to the need to ensure the smoothness of the video, video segmentation has higher requirements for real-time performance. In recent years, deep learning methods have developed by leaps and bounds. In terms of accuracy, deep learning methods have greatly improved compared with traditional methods. Therefore, image segmentation methods based on deep learning have gradually become a research hotspot. With the development o...

AI Technical Summary

Problems solved by technology

[0003] The present invention aims at the relatively low computing power of existing mobile devices and achieves the purpose of using an image segmentation method based on deep learning to display the results of each frame of image segmentation in real time in the application of video on mobile devices, and proposes a low-delay video Segmentation real-time preview method to solve problems existing in existing research topics

Images