Centroid of the image keeps affine covariance,but it can be only used to eliminate translation.In order to extract affine invariant features conveniently,it is necessary to construct centroid-like points that keep affine covariance and are different from the centroid.The existing algorithms are either computationally expensive or not suitable for binary images.In view of this,G-centroid is proposed.It was obtained by modifying the definition of centroid in the polar coordinate system,and the transformation function of integral along radial direction was introduced,thus making most of current centroid-like construction algorithms are special cases of the proposed G-centroid.Experimental results show that obtained G-centroid is more robustness to noise by introducing functions about central projection,compared with the generalized centroid,cross-weighted centroid and other methods.
Feature extractionAffine covarianceParametric estimation of affine transformationsCentroidRobustness to noise
万方数据
维普
