Segmentation of Human Instances Using Grab-cut and Active Shape Model Feedback
Esmaeil Pourjam, Ichiro Ide, Daisuke Deguchi, Hiroshi Murase · 2013
For a long time, image segmentation has been of great interest for researchers. Although there have been many studies on automatic object segmentation, still a method which can cope with arbitrary input situa-tions is out of reach. In this paper, we present “Ac-tive Shape Feedback Segmentation ” (ASFSeg) method, which is a way to automatically segment human sub-jects (more accurately, pedestrians) from images. For this task, we try to use masks generated by the Ac-tive Shape Model (ASM) algorithm as a prior input for the Grab-cut technique to segment the desired hu-man subject in the image without user interaction. To achieve this, we propose a feedback framework for the ASM sample generation. This feedback process com-pares the segmentation result mask from the Grab-cut stage and the samples generated in the ASM stage and then chooses the closest match to generate new sam-ples for segmentation. This process is repeated until the system converges on one of the generated masks. Some experiments are carried out that shows the validity of our proposed method, which shows that the system can automatically segment the human subjects in provided pictures. 1