The Study On Improved Hough Transform Based Algorithm

In this paper, the author introduces some research work on Hough transforms which is a powerful tool for finding basic features in digital images. Research work mainly included edge detection, line extraction, circle and ellipse extraction. The work is the research on basic theory. The motivation of this work is to find algorithms with high accuracy, high speed and robustness, so as to combine the application. The contributions of this work are listed below.Firstly, from the comparison of classical edge detection algorithms, find an optimize one with high speed and high precise. According to the information on edge grads direction, suggests a directional edge detection method as a good preparation.Aiming at existing defects of traditional Hough transform on line extraction, such as the imprecise, multi-peak detection and sensitive to noise, the author suggests progressive Hough transform algorithms.(1)An improved Hough transform based on peak detection by local operators, deduces the equation on the space division, adopt the dynamic sub-image feature to extract multi-peak. Not only wipes off the noise disturb, but also matches the precise and speed well. (2)An improved Hough transform based on two boundaries, experiment results shows multi-peak detection and the pretense peak can be dealt with well, the robustness improved obviously.Aiming at existing defects of traditional Hough transform on circle and ellipse extraction, such as large storage space consuming, time consuming, bad robustness, and the author suggests pre-processing the pixel with convexity, according to the convexity direction and directional edges to get hierarchical images. Then from the hierarchical images select points to solve the curve parameter, which can make the storage space less and have good robustness.At last, the author do comparisons with classical algorithms experiments, then analyze the evaluate targets.