Boundary-based corner detection using ANNs

Corners on a curve arise where two relatively straight line segments intersect. Corners in a graylevel image occur in regions of rapidly changing intensity levels and gradient directions. Therefore, corner detection techniques can be classified into two categories: binary and gray level. The binary approaches involve, first segmenting the image into meaningful regions, and then extracting boundaries from the regions of interest. Corners on the extracted boundaries are typically identified for those points with high curvature. Methods of discrete curvature measurement [7, 8, 9] are based on the rate of change of tangent direction. A variety of curvature measurement methods have also been reported by Teh and Chin [10]. The boundary-based corner detection methods [11 17] generally are easy to implement and computationally fast. However, when the curve is rotated in different orientations, they may suffer from poor detection due to the instability of discrete curvature measurement in digital images.