Affine Invariant Feature Extraction Based On Polar Radius Moment And Generalized Zero-order Momen

Affine invariant feature extraction plays an important role in pattern recognition and machine vision.Due to the convenience of image moment calculation,it has become a commonly used global invariant feature extraction technique.However,high-order moments are sensitive to noise,and only a limited number of invariants constructed from low-order moments are commonly used in practice.In order to improve the robustness of moment structure invariants,most of the current technologies are at the cost of increasing the amount of computation.At the same time,although the zero-order moment has strong robustness,it is not directly used in the construction of invariants.In order to construct affine invariants by using more low-order moments to enhance their robustness to noise,polar radius moments and generic zero-order moments are proposed.(1)Polar radius moments are proposed.It is actually the moment of the image formed by the algebraic sum of the gray value of the image on the symmetric polar radius,and the traditional image moment is only a special case of this polar radius moment.An algorithm for constructing affine invariants is given using polar radius moments.In particular,the first polar radius moment can be regarded as the generic zero-order moment,which can not only be used to normalize relative affine invariants to extract absolute affine invariant features,but also can be directly used to construct affine invariants.The experimental results also show that the moment constructed by the proposed algorithm is affine invariant,and the invariant constructed by the low-order polar radius moment has stronger robustness.(2)Generic zero-order moments are proposed.The zero-order moment is the strongest robustness,however it can only be used to normalize other relative invariants to achieve the construction of absolute invariants.This paper considers replacing the traditional zero-order moment with other more robust quantity for this normalization.To this end,looking at the zero-order moment from another angle,it is generalized as the area of the region enclosed by a class of closed curves with parameters,and is called generic zero-order moments.The relative affine invariance of generic zero-order moments are proved,and it is theoretically indicated that the noise immunity of generic zero-order moments is weaker with the increase of the order.Therefore,it is proposed to use the zero-degree generic zero-order moment instead of the zero-order moment to normalize the relative invariants,and use the zero-order moment to construct the affine invariant.The experimental results also show that the zero-degree generic zero-order moment is more robust,and the zero-order moment can also be used for the construction of affine invariants.