Affine Invariant Feature Extraction Based On Mellin Transform

The extraction of image affine invariant feature is widely used in computer vision.The affine invariant feature extraction plays a key role in many practical problems.Mellin transform has been widely used for the extraction of similarity invariant features,but similarity transform is only a special case of affine transform.This paper considers applying Mellin transform to the affine invariant feature extraction.The global affine invariant feature extraction algorithm can be divided into the contour algorithm and the region algorithm.The central projection transform can combine the contour and the region algorithm,but there is a loss of information along the polar direction of the target.The Mellin central projection transform is proposed in this paper and makes up for loss of information along the polar direction of the target with the central projection transformation.This paper uses Mellin central projection transform to construct affne invariant.Moment-based method is an important method of extracting affne invariant features in the region algorithm.However,the high order moment is sensitive to noise.Only a few integer order moments can be used to construct affine invariant in practice.Therefore,this paper proposes the Mellin integral moment based on the Mellin transform.Main results of this thesis are as follows:(1)The Mellin center projection transform is proposed,and the central projection transform is only a special case of the Mellin center projection transformation.A method is provided for the extraction of affine invariants by employing the area of closed curves derived by Mellin center projection transform.In fact,the area of the closed curve is the generalization of the traditional zero order moment,it is called the generic zero order moments.The experiment also shows that the affine invariants constructed by the generic zero order moments are more robust to noise than the affne invariants(AMIs)constructed by the traditional moments.(2)The Mellin integral moment is introduced and the order of traditional moment is generalized from integer to complex number.Method is provided for constructing affine invariant by Mellin integral moment.The traditional affine moment invariants(AMIs)can be derived in terms of Mellin integral moment.The results of experiment show that the affine invariants constructed by the real number order Mellin integral moments whose orders are less than 2 are more robust to noise than the affne invariant(AMIs)constructed by the traditional moment.