Randomized Kd-trees based region growing for segmentation on curve surface

A 3D model obtained from stereo cameras has a massive amount of data, and the structure of its points is so much simple with only the spatial information and colors. In this paper, a novel method of segmentation on the curve surface is proposed. Such segmentation method has applications in burn area estimation, texture mapping and local deformation. The segmentation method utilizes a novel region-growth algorithm combining randomized kd-trees algorithm for nearest-neighbors search. Moreover, the vector and distance information of the points is combined to detect the optimal control-points for approximate geodesics as the border of the region of interest. The segmentation method was evaluated on three models in different scale, and its effectiveness was verified on a real body model with patches. The experiments show its efficiency and utility.