A LINE SEGMENT DETECTION ALGORITHM BASED ON STATISTICAL ANALYSES OF QUANTIFIED DIRECTIONS IN DIGITAL IMAGE

Liang Jia, Nigang Sun

School of Information Science & Engineering, Chang Zhou University, 213164, China

Line segment detection is a typical image processing problem with constantly evolving solutions. Following the line segment detect (LSD) by Grompone von Gioi, two branches of algorithms merged. The first branch aimed to improve its speed at the cost of lower accuracy; the second applied in the opposite way. We investigated the philosophies of these methods, and attempt to develop a line segment detection algorithm based on statistical analyses of quantified directions (LSDSA) to achieve better accuracy and faster speed. We utilize a statistical approach estimating the distributions of pixels with direction values approximating the direction changes when traversing along the edges given by any edge detector. It efficiently reduces the dimension of the input data, and incurs limited increasing in computation time for validation process. The simulation results show that the proposed algorithm achieves better performance compared to the existing typical LSD algorithms. The experiment using industrial data in noisy cases also exhibits excellent performance.