Characteristic fusion method based on kernel typical correlation analysis

Abstract

The invention discloses a characteristic fusion method based on a kernel typical correlation analysis, comprising steps of using a kernel function to map an image matrix to a kennel space, extracting two groups of characteristic vectors of one mode, establishing a criterion function for describing the correlation between the two groups of the characteristic vectors, obtaining two groups of typical projection vector sets according to the criterion function, and extracting a combined typical correlation characteristic through a provided characteristic fusion strategy and applying in the classification identification. The beneficial effects of the characteristic fusion method are that: the kennel introduction technology successfully promotes the linear typical correlation analysis to the non-linear, can extract the characteristic having the stronger discriminability, avoids the decomposition on the matrix after mapping, and simplifies the data operation. The kennel typical correlation analysis algorithm delicately uses the correlation characteristic between the two groups of the vectors as effective determination information, which not only achieves the goal of information fusion, but also eliminates information redundancy between the characteristics and achieves a good result in the classification.

Description

technical field [0001] The invention relates to the field of feature fusion method, image processing, pattern recognition and artificial intelligence, in particular to a feature fusion method based on kernel canonical correlation analysis. Background technique [0002] With the development of computer technology, the feature-level fusion method based on information fusion technology has been applied in combined feature extraction, and has achieved good application results. The advantage of feature fusion is obvious, because the different feature vectors extracted from the same pattern reflect the different characteristics of the pattern, and the optimal combination of them not only retains the effective identification information of multiple sets of features participating in the fusion, but also eliminates the feature vector between redundant information. Existing feature fusion methods are mainly divided into serial fusion methods and parallel fusion methods. In recent ye...

AI Technical Summary

Problems solved by technology

[0004] The above classical statistical methods are all based on the assumed linear space, and cannot extract the nonlinear relationship between different features, and the kernel method is an effective way to solve the above problems, so it has become a trend to use the kernel to enhance the above statistical methods, thus deriving Kernel Principal Component Analysis (KPCA), Kernel Canonical Correlation Analysis (KCCA) and Kernel Partial Least Squares Analysis (KPLS) etc.

Images