Abstract:

The traditional Shi-Tomasi corner detection algorithm has been widely applied in many fields of computer vision. However， this algorithm has low efficiency and accuracy， poor noise resistance， and is prone to producing false corners. A method that combined adaptive threshold Canny edge detection and improved Shi-Tomasi corner detection was proposed. Firstly， improved Canny edge detection was used to extract image edges and screen candidate corner points， while a one-dimensional information entropy adaptive threshold was used to adapt to different image environments， thereby improving the efficiency and robustness of detection. Secondly， using circular templates for non-maximum suppression reduced the number of false corner points and enhanced the algorithm’s ability to recognize true corners. Finally， the improved Shi-Tomasi algorithm was applied to the extracted edge images for corner extraction， thereby achieving accurate image localization. The experimental results show that compared with the traditional algorithm， the proposed method has significant improvements in runtime and accuracy， and it has significant advantages in rotation invariance and noise resistance.

Key words: corner detection, adaptive threshold, Canny, Shi-Tomasi