In this paper we present an auto-detection cornerbased on eigenvalues product of covariance matrices (ADEPCM)of boundary points over multi-region of support.The algorithm starts with extracting the contour of anobject, and then computes the eigenvalues product ofcovariance matrices of this contour at various regions ofsupport. Finally determine automatically peaks of the graphof eigenvalues product function. We consider that pointscorresponding to peaks of eigenvalues product graph arereported as corners, which avoids human judgment andcurvature threshold settings. Experimental results show thatthe proposed method has more robustness for noise andvarious geometrical transform.
展开▼