A Corner Extraction Method Based on Multi-spectral Double-directional Detection and Multi-scale Corner-characters Validation

In this paper,TK(Tomasi-Kanade) corner-detector and COVPEX(Corner validation based on corner property extraction,COVPEX) algorithm are combined to extract corners from IKONOS Multi-spectral images.Firstly,we use TK corner-detector and find that the detector is sensitive to the corner-orientation and corner-contrast changes.It always results in "under-detection".Aim at these defects of TK corner-detector,we improve it and propose the Multi-spectral double-directional TK corner detector.To reduce sensitive degree to the corner-contrast,the new detector uses multi-spectral data to estimate corner-significance of a pixel.Moreover,the corner-clusters are presented in the processing results,which destroy uniqueness of corner.In order to reduce the proportion of pseudo-corners,we propose multi-scale COVPEX algorithm which uses multi-scale corner-characters to validate corners.In the meantime,based on the local minimum value theory,the corner-cluster removing method is also proposed to preserve uniqueness of corner.Finally,the comparative experiment results of corner extraction show that the proposed method is suitable for multi-spectral high-resolution imagery,accuracy and rationality of the extracted corners are considerably improved.