The GrabCut Segmentation Technique as Used in the Study of Tree Image Extraction

Tree image segmentation is an important algorithm and technique to separate an individual tree from the surrounding landscape in a photographic image. In recent years, this problem is a hot issue for scientific workers in forestry research field. In order to get a better processing result to the problem that a large number of holes and transparent phenomenon appear in the internal area of tree image, this paper makes use of GrabCut algorithm to extract the object tree. GrabCut is an innovative segmentation technique that uses both region and boundary information in order to perform segmentation. This approach based on optimization by graph-cut has been developed which combines both texture and edge information. The user interaction of this algorithm is friendly and the alpha-matte around an object tree boundary and colors of foreground pixels estimating is scientific. Therefore, it is suited for tree image extraction. The experimental results show that it is very effective when the background is not so complex. But when there are a lot of green plants in the background, the extraction result is not ideal. And because the delay due to the Graph Cut step is still noticeable, so the segmentation needs much computational time. Methods for incrementally updating the segmentation speed and accuracy could be explored in the future research.