Abstract:

To reduce the influence of edge tiny changes and noise on the corner detection and improve the detection accuracy, a new corner detection algorithm using multiscale autocorrelation matrix on edge contours is proposed. Firstly, the edge contour of an image is extracted by an edge detector. Meanwhile, the image is smoothed by the Gaussian kernel function with different scales. For each pixel on the contour, the variation information of gray magnitude on certain scales can be obtained by the differential operator and will be used to construct the autocorrelation matrix. Finally, the normalized eigenvalue of autocorrelation matrices on different scales is defined as the new corner measure. An edge pixel is specified as a corner when its measure is not only above the preset threshold but also the local maximum in a small region. Experimental results show that the proposed algorithm achieves better detection accuracy and fewer false corners.