Application of Support Vector Regression in the Detection of Image Geometric Feature

In this paper, a support vector regression (SVR) based method is proposed to detect a geometric feature such as line equation, corner point and angle degree between straight lines in an image. Digital image with geometric figures is collected and transmitted into computer. Median filter is used to reduce noise in the original gray scale image. Then image contour with single-pixel width is obtained by image segmentation. Regression function of each detected straight line is obtained by training the SVR with the training point set got from image contour. Then through calculating, we obtain corner points of the geometric figures and angle degree between every two straight lines, which have sub-pixel accuracy. Experimental results show that the proposed method is effective.