Extraction Of Topological Structure Of Plants Images

As plants is a universal natural entity, the three dimensional reconstruction of it has important significance in virtual reality, biology, landscape design and etc. In order to realize the three dimensional reconstruction of plant skeleton, the topological structure should be extracted from plant images firstly. Referring to the low automation computing in topological structure extraction at present, in this paper, we study and realize the extraction of topological structure of plants from images and the feature matching of topological structure. Our main research contents are as follows:(1)By using GrabCut algorithm, based on interactive image segmentation, the segmentation problem under complex background of natural plant images is solved. First, traditional image segmentation method and digital matting method have been used to divide the background of plant images, and the segmentation is unsatisfactory. Then, as an interactive segmentation method, the GrabCut algorithm has less interaction, we choose GrabCut algorithm to realize the segmentation of plant images. Besides, Berkeley segmentation dataset and plant images are used to test the algorithm. The experiment shows that the algorithm can segment the background of images accurately, and then achieve the segmentation of complex background of plant images.(2)We propose an extraction method of plant topological structure, and obtain single pixel plant skeleton with relatively complete structure. After segmenting background of plant images, firstly, we get the plant contour by image pre-processing and contour tracking. Then, interval sampling pixel method is used to hash the plant contour, and then triangulate the plant contour by Delaunay triangulation algorithm. Finally, we calculate the triangle cores and use the ray method to remove redundant triangle cores which are out of plant contour, and construct the plant topological structure with the triangle cores. In order to improve the algorithm speed and maintain the integrity of the plant topological structure, we compare different degrees of plant contour hash results, and choose interval 25 pixels to hash the contour. Four representatives kinds of plants are selected to extract topological structure in the experiments, and all of them have achieved the expected results.(3)We propose an improved bidirectional SIFT feature matching algorithm which can be used to achieve feature matching of plant topological structure. To remove the mismatching, two steps are to follow. First, SIFT bidirectional matching algorithm is taken to eliminate part of the mismatch. Second, disparity gradient constraint and RANSAC algorithm are used to purify the matching points. In speed improving, not only K nearest method is taken at the beginning of matching, but disparity gradient constraint is optimized. They both reduce the iterations to lower the time consuming. Experiments show that the improved algorithm can exclude about 35%~65% of matching points. Furthermore, compared with SIFT algorithm, the time consuming of our algorithm can reduce by 4%~10%.