Research On Brain Image Recognition Algorithm Based On Tensor Decomposition

With the rapid development of modern computer technology, the medical image research has entered the stage of computer-aided detection and diagnosis. The diseases of brain tumors,Alzheimer’s disease and Parkinson’s disease can be detected by the advanced medical equipment.The brain images can be recognized and analyzed by the image processing technology. This can help doctors improve the accuracy and reliability of brain disease diagnosis. Medical image processing has an important reference value in clinical diagnosis.Most images have a natural tensor structure, or can be organized into a tensor structure. The medical images are no exception. A tensor has a strong ability for computing and expressing. It can be used to represent the linear relationship between a scalar, vector and tensor. It can maintain the structure information when the images are transformed into the form of tensors. It provides a basis for the analysis of high dimensional data. The two algorithms of Fourier higher order singular value decomposition and wavelet high order singular value decomposition are presented in this paper, according to the structure characteristics of tensor decomposition algorithm. Next the two algorithms are used to identify the brain tumor images, respectively. The main work is as follows:(1) First briefly summarizes the research status of modern medical image processing technology. It finds that the experiment result for modern medical image processing method is not ideal for brain image by doing a lot of analysis and comparison. In this paper, tensor decomposition algorithm is applied to the image processing of the brain images, according to combine with the characteristics of the brain images itself and the study of the tensor decomposition algorithm theory. It has a very important research value.(2) An algorithm of brain image lesion recognition based on Fourier higher order singular value decomposition is presented. The algorithm combines Fourier transform with higher order singular value decomposition. It will proceed Fourier transform before calculates higher singular value. The algorithm has some good properties in dealing with the stationary signal by comparing with the traditional higher order singular value decomposition. And the algorithm doesn’t need to be repeated choose the minimum root operation in the middle of calculating.Because the algorithm is drove by data, it can extract the structure feature of multidimensional data group with doesn’t need to select parameters or set the threshold. The algorithm can identify health and pathological changes brain images quickly and accurately.(3) An algorithm of brain image lesion recognition based on wavelet higher order singularvalue decomposition is presented. The algorithm combines wavelet transform with higher order singular value decomposition. It will proceed wavelet transform before calculates higher singular value. The algorithm has a special advantage in dealing with the non-stationary signal by comparing with the algorithm of brain image lesion recognition based on Fourier higher order singular value decomposition. It solves a lot of problems while Fourier higher order singular value decomposition algorithm can’t solve.The two algorithms of brain image lesions recognition based on tensor decomposition are presented in this paper. The simulation results show that these two algorithms are suitable for dealing with brain images. The health and pathological changes of brain images can be identified by the two algorithms quickly and accurately. They have a very high research value.