Research On Target Feature Data Fusion And Recognition Technology In Multi Source Vision Scene

With the development of science and technology,especially the computer image information processing,more and more image information needs us to identify and classify.The traditional human identification and classification is not only to take up a lot of manpower and the recognition efficiency is low,so the image feature level fusion technology.Image feature level fusion is a dimension reduction operation which is based on the premise of image feature extraction.Easy to follow the identification of the operation,so that the recognition speed is faster,more efficient.If the two sensor receives the infrared image and visible light image,the two sensor produces different source images of the same scene,to identify the judge if the traditional method is two times more than the number of sources,and with the increase linearly,and the image features of data fusion integrated features two images the value of fusion in comparison,effectively remove redundant and improve the recognition efficiency.In view of the feature level fusion of the advantages of image processing,this paper of the same scene images of different source of image preprocessing such as Gauss filtering and median filtering method may cause noise suppression of the second,and on the preprocessed image feature extraction Hu moment invariant moments,gray co-occurrence matrix,affine invariant moment and wavelet moment invariants.And the invariance and stability of the characteristic moments are analyzed and used to feature data fusion.Then the data fusion,the typical covariance matrix fusion,principal component analysis and fusion,based on the traditional multivariate statistical analysis method in the fusion algorithm in the sample number must be greater than the number of variables of the shortcomings,and propose a new multi feature fusion based on partial least square method,further redundancy removal and use of two classification algorithms to improve the traditional SVM in the recognition process to carry out multi classification without sacrificing the recognition rate under the condition of.Finally,the experimental results show that the feature fusion algorithm can not only remove the redundancy effectively when dealing with multi feature fusion,and has a high recognition rate in the subsequent multi class recognition process.