Self-adaptive weight multi-view discrimination method

Abstract

The invention discloses an adaptive weight multi-view discrimination method. The method comprises the following steps of constructing undirected weight maps of different views of a data set, and calculating a Laplacian matrix Ls; carrying out consistency constraint on different view data on the basis of a Hilbert-Schmidt independence criterion (Hilbert-Schmidt Index Criteria, HSIC), and calculating a constraint matrix T; optimizing the projection matrix P in combination with a consensus low-rank sparse representation learning method; a weight parameter being introduced, and a corresponding weight being adaptively given according to the information amount contained in each piece of view data; constructing a final multi-view discrimination model of the adaptive weight; solving an optimal multi-view projection matrix of the model by solving the target model; and discriminant analysis being carried out on the test set samples, and the accuracy of image recognition being obtained by using a KNN algorithm. According to the method, the consistent structure of different views can be kept for the image data with noise pollution, and the method has very high accuracy and robustness.

Description

technical field [0001] The present invention relates to the technical field of image recognition, and more specifically, to a multi-view discrimination method with adaptive weights. Background technique [0002] With the continuous advancement of technology, the acquired original image data is usually diverse and high-dimensional, which will lead to great challenges in processing multi-view data. In order to reduce the consumption of computer memory and the amount of data calculation, the two most classic optimization algorithms, principal component analysis (PCA) and linear discriminant analysis (LDA) algorithms, will be considered first. However, since the above algorithm is sensitive to noise, He et al. proposed a locality preserving projection in their paper "Locality Preserving Projections", which improves the robustness of the algorithm to noise by maintaining the local structure information of the data. [0003] Compared with single-view learning algorithms, multi-vi...

AI Technical Summary

Problems solved by technology

[0005] The present invention provides a novel multi-view discrimination method with self-adaptive weight in order to solve the unsatisfactory discrimination effect in the case of multiple different views, high dimensions and serious noise pollution in the prior art.

Images