Video segmentation improvement method based on OSVOS

Abstract

Aiming at the problems of segmentation target missing and inaccurate target marking caused by factors such as illumination and color in OSVOS, the invention provides a video segmentation improvement method based on OSVOS, and the method comprises the steps: calculating the color component of a video image under the morphological theory, comparing the color component with a standard library, and dividing the video image into a strong light condition and a weak light condition. Meanwhile, the color similarity is calculated; after being compared with a threshold value, the data is divided into asimilar condition and a dissimilar condition, wherein the two influence factors correspond to four conditions, and different processing methods are adopted for the four conditions, including hole filling of a target missing part of an OSVOS segmentation result and removal of a non-target connected region, so that the quality and accuracy of the segmentation result are improved.

Description

technical field [0001] The invention relates to the field of computer vision and underlying video segmentation technology, in particular to an OSVOS-based video segmentation improvement method. Background technique [0002] The development of deep learning has promoted the development of computer vision and successfully improved the development of related applications such as object recognition, video detection, video segmentation, etc., especially video segmentation technology. Video segmentation is widely used in medical imaging, intelligent traffic management, network education, video surveillance and other fields in people's daily life. At the same time, the development of neural network further improves the performance of target segmentation technology. Video segmentation techniques can be roughly divided into two categories, semi-supervised and unsupervised segmentation methods according to different initialization methods. Semi-supervised segmentation refers to mark...

AI Technical Summary

Problems solved by technology

OSVOS regards the video as a still image and separates it frame by frame. This feature determines that it can complete a good segmentation when the lighting is normal and the color difference between the target and the background is obvious. However, when the light is too strong, there will be large The color of the range is distorted, and bright spots appear locally; when the light is too weak, the overall will be dark, resulting in blurred target objects and backgrounds, and prone to missing some areas

On the other hand, when the color of the target object is similar to the color of the background image, the OSVOS algorithm will mistakenly judge part of the background image as the target, so that there will be more non-target connected regions in the result map, which will cause the degradation of segmentation quality.

Images