Robust canonical correlation analysis algorithm based on generalized mean value

Abstract

The invention discloses a robust canonical correlation analysis algorithm based on a generalized mean value, and mainly solves problems that a conventional canonical correlation analysis algorithm is not robust for outliers and a high-dimensional small sample causes a singular problem of a sample covariance. The implementation process comprises the steps: (1), inputting necessary parameters, and carrying out the centralized processing of a training sample; (2), solving two groups of projection sets of conventional canonical correlation analysis; (3), reconstructing a target optimization function of a model based on a generalized mean value, so as to inhibit the impact on a target function from the outliers; (4), solving the target function through employing a linear iteration method, wherein the two groups of projection sets of conventional canonical correlation analysis are used for initialization; (5), enabling the solved two groups of projection sets of the conventional canonical correlation analysis based on the generalized mean value to be used for the feature extraction and dimension reduction of the sample. The experiment results on a multi-feature handwritten form database (MFD), a human face database (ORL) and an object image database (COIL-20) verify the effectiveness of the algorithm.

Description

technical field [0001] The invention belongs to the technical field of feature extraction and data dimension reduction, and mainly relates to the improvement and optimization of typical correlation analysis algorithms. Specifically, it is a robust canonical correlation analysis algorithm based on generalized mean, which can be applied in fields such as machine learning, pattern recognition, data mining and image processing. Background technique [0002] In the field of pattern recognition and machine learning, dimensionality reduction (DR) has always been one of the hot research topics. A large number of methods have been proposed for data dimensionality reduction, among which Principal Component Analysis (PCA) is one of the most classic methods. But PCA focuses on feature extraction and dimensionality reduction of single-mode data (Single-viewdata). With the development of science and technology, the means for people to collect data are more diversified, and the same thin...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem of non-robustness to outliers and high-dimensional small samples commonly existing in typical correlation algorithms such as traditio

Images