Corner Characterization by Statistical Analysis
of Gradient-direction
Shi Yin
Department of Industrial Engineering
University of Toronto
Toronto, Ontario, Canada M5S 3G9
yin@argos.rose.utoronto.ca
Jens G. Balchen
Department of Engineering Cybernetics
Norwegian University of Science and Technology
N-7034 TRONDHEIM, NORWAY
jgb@itk.unit.no
Abstract
A new approach to corner (junction) characterization is proposed in this paper. A corner in grey-level images is characterized by the number of lines and their orientations which construct the corner. Therefore a corner can be classified as an L junction, a T junction, or a Y junction. In the proposed framework, the corner location (intersection point of the lines) can also be detected in subpixel accuracy. Our approach is based on the statistical analysis of gradient-direction in an intensity image, and takes the signal-to-noise ratio (SNR) into account. This paper also discusses some practical aspects of this approach, such as computational complexity and the scale-space problem. Experiments with both synthetic and real images reveal that the proposed approach can cope with noise much better than other techniques.
This paper is currently consitional accepted by IEEE Transaction on Image
Processing. The
full paper in postscript format is available via ftp.