Abstract:

This paper presents a novel method of 3D reconstruction from calibrated images. With this method, firstly, initial corresponding points are found by using SIFT, secondly, 3D points are generated as seed points, thirdly, region growing is implemented according to these seed points until the whole surface is reconstructed. During the process of region growing, image correlation of the projected patches in two images is calculated. If the image correlation is bigger than a threshold, the growing is considered as a right one; otherwise the growing will be rejected. The performance and effectiveness are evaluated by experiments.

Key words: surface, seed points, region growing, image correlation, 3D reconstruction