J4 ›› 2013, Vol. 40 ›› Issue (4): 119-124+129.doi: 10.3969/j.issn.1001-2400.2013.04.020

• Original Articles • Previous Articles     Next Articles

Corner detection via anisotropic Gaussian kernels information entropy on edge contours

ZHANG Weichuan;SHUI Penglang;XU Guojing   

  1.  (National Key Lab. of Radar Signal Processing, Xidian Univ., Xi'an  710071, China)
  • Received:2012-05-03 Online:2013-08-20 Published:2013-10-10
  • Contact: ZHANG Weichuan E-mail:zwc2003@163.com

Abstract:

To improve the accuracy of corner detection, a new corner detector is proposed based upon the information entropy which is derived by the anisotropic Gaussian directional derivatives (ANDDs) on edge contours of an image. Firstly, the edge map of an image is extracted by the Canny edge detector. Secondly, small gaps between contours are filled. Finally, on each contour pixel, the main direction corresponding to the maximal ANDDs at each contour pixel and its surrounding pixels are used to compute the main direction's probability density function and information entropy. Different from the traditional contour-based detectors, our detector uses the maximal intensity variation's directional information entropy on contours and surrounding pixels rather than the curvatures of the planar curves, which presents better robustness. Experimental results show that the proposed detector achieves a better corner detection performance than several state-of-the-art detectors.

Key words: edge contour, anisotropic Gaussian kernels, information entropy, corner detection