Method for rapid cooperation expression of face classification

Abstract

The present invention discloses a method for rapid cooperation expression of face classification. The method comprises: obtaining a plurality of face images of all the measured personnel, and combining a training sample set and a test sample set; projecting the training sampling set into a PCA subspace, and obtaining the feature face set of the sample set; performing common linear expression of the test samples through adoption of the feature faces and an intra-class variable dictionary; initializing dictionary coefficients through the 12 paradigm iterative computations in a dictionary space, and subsequently performing second introduction of 11 paradigm to complete the accurate optimization of the coefficients; projecting reconstructed test samples PCA coefficients and dimensionality reduction samples to an LDA subspace, and respectively obtaining LDA coefficients of the reconstructed test samples and the training samples; and selecting the category having the minimum reconstruction errors with the test samples as the final candidate category of face test samples. The method for rapid cooperation expression of face classification effectively rejects the information redundancy between the training samples and the test samples, improves the face identification precision, greatly reduces the time expenditure of a traditional expression optimization method and has good universality and the robustness.

Description

technical field [0001] The invention discloses a fast cooperative representation face classification method, which involves representation learning, dictionary optimization and data dimension reduction technology. Specifically, it includes the construction of intra-class variable dictionary, PCA dimension reduction, quadratic sparse collaborative optimization, and LDA identification and reconstruction error evaluation; it belongs to the technical field of image processing and pattern recognition. Background technique [0002] Face recognition (FR) is one of the hottest research areas in pattern recognition, computer vision, and biometrics. But it is often affected by many factors, such as posture, expression, lighting and occlusion. In addition, not all collected high-dimensional data are effective for classification. In order to obtain efficient face recognition performance, researchers have proposed various classic dimensionality reduction methods in recent years, such as...

AI Technical Summary

Problems solved by technology

[0006] In the case of insufficient training samples, the traditional representation learning optimization method ignores the information redundancy and lack of discrimination between samples in the process of reconstructing high-dimensional original data, and at the same time avoids the use of l 1 Due to the huge time overhead caused by the paradigm solving minimization problem, a fast collaborative face classification method is proposed. After the data dimensionality reduction model is established, the method is more robust by designing a secondary optimization strategy based on mixed paradigms. The sparse coefficient vector of , thus speeding up the optimization and improving the accuracy of face recognition

Images