Image Analysis Based On Discrete Orthogonal Moments

With the rapid development of social informatization technology,the role of the information contained in the image is more and more important.People can get information you want with the suitable image processing.Image moment vector is a global product which can describe the image features,while image moment invariant still maintain good stability with geometric changes.Besides,image moment invariant has a good robustness to noise.They are widely used in image processing fields.The image processing based on discrete orthogonal moments is used as global feature descriptors,the analysis and classification of image in this paper.The main research work and innovation points are as follows:(1)On the background of image moment development,the geometric moments,geometric invariants,Krawtchouk moments,Tchebichef moments and Hahn moments are introduced in detail in this paper.Besides,the quick calculation method of moments is presented.The corresponding invariants of discrete orthogonal moments is deduced based on a linear combination with geometric moment.We also analyze experimentally Krawtchouk moments,Tchebichef moments and Hahn moments in terms of image reconstruction,image classification and recognition results.(2)Based on two kinds of polynomials which are different from traditional discrete polynomials,we propose a novel kind of image discrete orthogonal moments-Charlier and Meixner moments.This paper analyzes experimentally these two kinds of discrete orthogonal moments under the subspace in terms of representation ability of image.With the same order,the results of image reconstruction of Charlier and Meixner moments are worse compared with the traditional orthogonal moments.Based on the orthogonal properties of polynomials,this paper proposes a new kind of discrete orthogonal moments-Krawtchouk-Meixner(K-M).We analyze experimentally this two-dimensional discrete orthogonal moments in terms of image reconstruction in different subspace,and the comparison with Tchebichef,Krawtchouk,and Hahn moments.The results show that the performance of Krawtchouk-Meixner moment works well.(3)The existing methods for extracting the translation and scale invariants from the discrete orthogonal moments are via a linear combination of the corresponding invariants of geometric moments or image normalization prior to moment calculations,which led to calculational errors and degraded their representation ability.In this paper,a novel kind of discrete orthogonal moments named as Charlier moment is proposed based on the discrete Charlier polynomials,and then an approach to directly derive the translation and scale invariants from Charlier moments is also presented.Experimental results show the high classification and representation accuracy of these invariants as a result of direct calculation instead of the image normalization or a linear combination of the corresponding invariants of geometric moments.It is also shown that these invariants are relatively robust in the presence of image noise and are potentially useful as a kind of invariant descriptors in many image analysis,pattern recognition and computer vision applications.