GrabCut Algorithm Fusion of Extreme Point Features
Chao Li, Xiaozhen Zhao, Huiying Ru · 2021
Image segmentation is an important part of computer vision. Among them, interactive image segmentation only provides useful prior knowledge to the computer. GrabCut interactive image segmentation is a classic algorithm for the use of graffiti, line drawing, and square in the user interaction process. The box interaction method brings inconvenience to the user's operation. Aiming at the above problems, a GrabCut interactive image segmentation algorithm fused with extreme point features is proposed. First, the user marks the top, bottom, left, and right extreme points of the target area in the image, and uses the algorithm to solve the extreme value box with the extreme point as the apex; then select the extreme value box as the foreground image, the image outside the extreme value box is the background image. Then the initial segmentation foreground and Gaussian mixture model parameters of Grabcut are determined, the foreground image is segmented using the minimum cut algorithm, and the energy function is minimized through several iterations to complete the target segmentation. Introduction.