Uncertainty in matching straight line segments

The fragmentation, incomplete extraction and excessive extraction are three kinds of common uncertainty problems in extracting straight line segments, which are critical to matching straight line segments across different images. Based on the analysis of the process of extracting straight line segments, a method is proposed to describe quantitatively the above two problems by the result of edge detection. With the positions and the gradients in gray of edge points close to the extension of straight line segments, the probability in position of endpoints of each straight line segment is defined. According to the above probability expression, the credibility of grouping different lines is measured and furthermore the measure that any two straight line segments extracted from different images satisfy the epipolar constraint and photometric similarity constraint is computed, with which the strength of each potential straight line segment matching pair is given. The experimental results on real stereo remote sensing images show that the method proposed in this paper is competitive.