Application Of Tensor Principal Component Analysis Algorithm In Brain Medical Images

With the rapid development of computer vision,related disciplines such as image processing technology and machine learning are also growing.With the successful application of medical images in clinical practice,medical image processing and classification work has become increasingly important.The classification of images is to distinguish different categories according to the different characteristics reflected by the pixels in different areas of the image.The classified images can be widely applied to various occasions,such as face recognition,hyperspectral remote sensing,computer guided surgery,and the like.The identification and analysis of brain medical images is a multidisciplinary and crossover hot development direction.The reason lies in the complexity of the human brain's organizational structure.The brain constitutes a three-dimensional spatial image data set with a large amount of data and limited sample marks,which increases the difficulty of brain image classification and recognition.Therefore,as one of the key techniques for brain image analysis and processing,the research on the feature extraction and classification methods of brain images has attracted more and more attention from domestic and foreign researchers,and has also become the focus of research in many fields.In solving the problem of feature extraction and classification of high-dimensional images,the average researcher is accustomed to solve it from the perspective of vector,and thus neglects the characteristics of the image organization,thus destroying the organization-related information of high-order images.In order to solve this problem and improve the performance of brain image classification accuracy,this paper focuses on the feature extraction and supervised classification methods of brain MRI images.Based on the tensor matrix,the principal component analysis algorithm of traditional image feature extraction is analyzed and improved,and it is extended to high-order tensor principal component analysis algorithm.The traditional principal component analysis algorithm is only the tensor principal component analysis algorithm.In a special case,the new algorithm has potential advantages over traditional methods.The new algorithm has more extensive significance,and it has advantages in computational efficiency and feature extraction.This article combines Brain Web:Simulated Brain Phantom Database(abbreviated as BPD)3D brain data sets on MRI brain image,and uses the domain characteristics of brain image of magnetic resonance image to map the original data of image pixels according to the definition and formula of data quantization.The quantification of the data was performed,and the features of the MRI brain image were extracted and classified using the tensor principal component analysis algorithm.For the traditional non-tensor classification requirements,the extracted high-order image feature data was sliced.Second,because the computation between the quantized data is performed within the framework of the tensor matrix of the defined circular convolution,the computational complexity in the real-number domain is particularly complex and requires a large number of calculations,algorithm for quickly calculating tensor principal components.In this paper,the quantized data set is directly brought into the two-dimensional fast Fourier domain,which greatly shortens the calculation time required for calculation.In the work of this paper,a large number of MRI Phantom data from a large number of MRI brain images were extensively tested using the new algorithm,and achieved good results.Experiments show that the tensor principal component analysis algorithm based on the tensor matrix framework in the image feature extraction And the classification effect is superior to the classic principal component analysis algorithm.The tensor model in this paper is different from the traditional mature tensor model.The tensor model expands the "t-product" tensor model in recent years,and the mathematical model of "tensor matrix" is obtained.The tensor matrix model preserves the dual characteristics of the rows and columns of traditional matrices,and is relatively well compatible with traditional matrix models.The work of this paper will use the theoretical results of the tensor matrix in the analysis and processing of medical nuclear magnetic resonance images.It has good application value and guidance significance.