Feature Fusion Based On Principal Component Analysis And Its Applications

How to get the information and use efficient ways for fusion is an important topic in pattern recognition because there is a lot of information in images. The feature fusion of image has more and more important applications, such as in target recognition, medical treatment and biology feature recognition. This dissertation mainly studied the image feature fusion based on PCA(principal component analyses) and its application of Small-weak target matching.Firstly, based on the character of Small-weak target image, this dissertation analyzed and chose multi-features which include fractal feature, multi-direction and multi-scale gradient feature, energy feature, mean gray feature, morphology feature and clustering feature. And moreover this dissertation proposed local gray probability feature, used the feature fusion algorithm based on PCA to fusion, and matched the target.Secondly, this dissertation reviewed the existing state of feature fusion, and introduced the present status of PCA. We used PCA, 2DPCA(two dimension principal component analyses), 2D~2DPCA(two directions of 2DPCA) and DiagPCA(diagonal PCA) in feature fusion, compared and analyzed the results of Small-weak target matching from these algorithms. This dissertation pointed out that the efficiency of algorithm based on DiagPCA was best, and the computational time of algorithm based on 2D~2DPCA was shortest.Lastly, we studied the computation efficiency of feature fusion algorithm based on PCA. Combined the Schmidt's orthogonalization algorithm, 2DPCA and DiagPCA to propose two improved algorithms such as 2DFPCA(fast 2DPCA) and FDiagPCA(fast DiagPCA). Applied these new algorithms to Small-weak target matching, and analyzed the computational time with different compression ratio.