Dense matching for wide baseline images based on equal proportion of triangulation

Abstract

The dense matching of pairs is the basis of advanced image processing applications, such as 3D reconstruction and simultaneous localisation and mapping (SLAM). However, it is difficult to obtain efficient dense matching results in wide baseline image pairs due to various conditions. For this reason, the authors propose a dense matching method for wide baseline images based on an equal proportion of triangulation. Firstly, this Letter proves the mathematical property of equal proportion of triangulation affine transformation, and then builds a triangulation based on the set of sparse matching points. By using this property, the positions of equal proportion points in the corresponding triangulation of the two images to be matched are calculated, respectively. Secondly, to eliminate the points and obtain the inner point set, these proportional points are used to check the similarity of the matching triangle regions. Finally, the inner point set is used to calculate the positions of the dense matching points as the final dense matching result. The experimental results verified the validity and accuracy of the proposed method and this method is suitable for tilt photography, multispectral and multidate remote sensing image dense registration, and more.