Line segments detection with scale analysis: A principal component analysis based approach

Hough transform and its revised versions are popular and powerful methods for line detection. However, most of them need edge images as input, and do not consider the scale selection. Besides, it is difficult for Hough-based methods to eliminate spurious peaks caused by noise points. This paper proposes a Principal Component Analysis (PCA) based approach for line segment detection. In the proposed algorithm, PCA is introduced for two purposes: to estimate the existence of line segment in a local region, and to extract line segment. The proposed approach mainly contains four steps: First, we give a multi-scale representation for the original image, thus a set of images with different scale is obtained; second, for each scale image, the gradient magnitude image is calculated by performing gradient operator. Then each gradient magnitude image is divided into overlapped windows respectively; third, for each window, we choose several representative points according to the gradient magnitude, and then perform PCA on these points; fourth, line segments are grouped according to PCA results and multi-scale analysis.