Background Removal Based on Significant Contour Tracking and Segmentation
Download as PDF
DOI: 10.23977/AICT2020004
Corresponding Author
ABSTRACT
The grabcut algorithm is known to use the graph theory to cluster pixels for background removal, however, it is disadvantageous that the mark is needed to identify foreground and background. To automate this process, a method is proposed which is consist of three parts. First the edge detection with max pooling is used to get foreground edges, and then to get the object-level contours, a significant contour tracking algorithm is used. When the relative precise contour is obtained, an algorithm fusing morphological operation and grabcut is applied to get the exact segmentation.
KEYWORDS
Max pooling; grabcut; background removal; morphology transformation