Automatic Matching in Convergent Close Range Image

Image matching is an essential task in digital photogrammetry. The fully aoutomatic matching of convergent close range image is a difficult task due to possible significant geometric distortions in these images. In this paper an efficient and fully aoutomatic matching method for convergent close range images is presented. The proposed method consists of three main steps. In first, the initial feature extraction and matching is performed using ASIFT (Affine-SIFT) algorithm. In the second step the epipolar geometry based on fundamental matrix is estimated using genetic algoritm to eliminate most of outliers in initial match features. Due to the defect of epipolar geometry to remove all outliers, in third step of proposed method a new approach to remove reminding outlier from previous step is proposed. In this method cluetering process is performed by k-Means algorithm and followed by a consistency check in the local affine transformation model for each cluster. The experiment results using both synthetic data and various sets of convergent close-range images show that the proposed algorithm provides reliabile and acuurate matching.