A target recognition of wavelet neural network based on relative moment features

In target recognition, the image of the object consists of the translation, rotation and scaling. Because the moment has some invariants, it has been used widely in the domain of image identification. The quadrature moment and the wavelet moment has the rotation invariant, but they must pass the normalization pretreatment talent to the translation and scaling of the target image. Hu invariant moments although have the rotation, translation and scaling invariant features, full the sensitive with error to the small target distortion and error because of its high order moment, limiting its application. In this paper, we use a new kind of moment algorithm to get some real rotation, translation and scaling invariant features of the target. Furthermore, we use wavelet neural network to classify target so that the flexibility and efficiency of the identification algorithm can be improved.