ROI feature extraction algorithm based on the multi-scale corner

With several extraction methods discussed focused on contour corner, this paper presented a detection method based on multi-scale curvature polynomial (MCP), by making use of multi-scale information and the scale uncertainty of corner. Region profile at different scales and the corresponding curvature values were obtained by using the detection method of the curvature scale space (CSS), and the weighted sums were used in the curvature corresponding to local maxima points on different scales. And for non-polar curvature point, the curvature product was adopted, all the curvature polynomial value of the region was calculated and the contour corner of the region was extracted by taking the threshold. The test shows that the algorithm is simple, and it is more effective in the extraction of region of interest (ROI) than the traditional ones.