Image Analysis And Recognition Based On Orthogonal Polynomial Transformation

Orthogonal polynomial transform plays an important role in image processing and machine vision.It can be applied to many fields including texture retrieval,image analysis,Image recognition,digital watermarking,template matching and edge detection,image compression and so on.At present the studies and applications of orthogonal polynomial transformation research are aimed at a specific polynomials,this is because different kinds of orthogonal polynomials are orthogonal in different coordinate system.Moreover,the existing orthogonal transforms are restricted to integer order,the "spectral" space of the orthogonal polynomial transform is not complete.Little investigation on non-integer order orthogonal transform,and its properties as well as applications have been conducted to date.In addition,invariant feature based on orthogonal polynomial transform is a kind of global description method.It can effectively construct images' rotation,scale,translation,affine and projection transform invariant features as well as has high noise robustness.Moreover,it is easy to realize blurring invariance.However,the recognition ability descrease sharply when image is occluded and the background is complex.This influent the application of orthogonal polynomial transform in image recognition seriously.Regarding the above issues,the fractional orthogonal polynomial transformation model and the local feature construction method of image based on orthogonal polynomial transform are studied.The research points are listed as followings:1.In this thesis,the general framework of fractional-order orthogonal polynomial transform which defined under both Cartesian coordinate system and polar coordinate system is proposed.Shifted Legendre polynomials are implemented in this thesis to investigate the properties of fractional-order orthogonal polynomial transform.Theoretical analysis and experimental results demonstrates that fractional-order orthogonal polynomial transform is not only inherits the properties of traditional orthogonal polynomial transformations but also is capable of region-of-interest(ROI)feature extraction.Meanwhile,it has potential for image reconstruction and face recognition and have high noise robustness in invariant image recognition.2.Interest point detection and region description algorithm based on the local Zernike polynomial transform is proposed.By constructing a response matrix based on the local Zernike polynomial transform which is similar to the Hessian matrix of the Speeded-Up Robust Features(SURF)algorithm.This thesis uses the matrix as the interest point detector and then Haar wavelet transform is used for describing the interest points.In the experiments part,this thesis compares the proposed algorithm with conventional algorithm-SURF.There are five kinds of contrast experiments under the condition of zoom and rotation,the change of perspective,image blur,JPGE image compression,light conditions changes respectively.Through the experiments,we can see that the orthogonal polynomial transform have better interest point detection performance than the SURF when in different rotation angle,the change of perspective and image compression.And the performance is similar to the SURF under the condition of image blur,and different light degradation.The experimental results show the effectiveness of our proposed algorithm.