Joint low-rank constrained cross-view discriminative subspace learning method and device

Abstract

Embodiments of the present invention provide a joint low-rank constrained cross-view discriminative subspace learning method and device. The above method includes: defining the objective function of the double low-rank discriminant subspace learning formula; using a supervised regularization item as a strong constraint condition, and re-elaborating the objective function; adding a joint heterogeneous regularization item to re-elaborate the objective function; image data The set is divided into a test set and a training set; use the training set to solve the value of each variable when the objective function value is minimized; obtain the characteristic subspace after solving the objective function; project the test set through the characteristic subspace, and obtain all category images in the data set All the features of the classifier are finally used to obtain the recognition rate of the data set. The invention combines the heterogeneous regularization term as a constraint to construct the discriminant term for feature learning, which can project the isomorphic and heterogeneous information of the sample into the subspace for the discriminant learning model of image recognition and classification tasks, and promote Model adaptability and robustness.

Description

technical field [0001] Embodiments of the present invention relate to the field of image classification, and more specifically, embodiments of the present invention relate to a method and device for learning a cross-view discriminant subspace with joint low-rank constraints. Background technique [0002] Cross-view learning has attracted much attention in recent years, since our images are often acquired from various angles or from different sensor devices. In recent years, many cross-view discriminative subspace learning methods have been proposed, which have not only attracted much attention but also been successfully applied in real work. However, the discriminative models constructed by these methods only separate the samples of the same view and different categories, and the samples of different views of the same category are close to each other, ignoring the isomorphic and heterogeneous information hidden in different views. Contents of the invention [0003] In thi...

AI Technical Summary

Problems solved by technology

However, the discriminative models constructed by these methods only separate the samples of the same view and different categories, and the samples of different views of the same category are close to each other, ignoring the isomorphic and heterogeneous information hidden in different views.

Images