Transductive low-rank tensor discrimination analysis method

Abstract

The invention discloses a transductive low-rank tensor discrimination analysis method, which comprises the following steps that: N image sets are given, and image set data matrix which is greater than1 and less than N in the data matrixes of the N image sets gives category labeling information, and N feature matrixes extracted by the N image sets are mapped into points on a Grassmann manifold; each point on the Grassmann manifold which represents an image set feature matrix is mapped to a symmetric space to form N b-order symmetric matrices; the N b-order symmetric matrices are combined to form a tensor; a target function is constructed for solving the discrimination low-rank representation matrix of the tensor; and an iterative convergence threshold value algorithm is adopted to solve the target function to obtain the discrimination low-rank representation matrix of the tensor. By use of the method, a situation that original image set data is directly taken as the input tensor is avoided, and calculation is carried out through transformation to Euclidean space through one piece of mapping on the basis of similarity between points on the Grassmann manifold so as to form the tensor.

Description

technical field [0001] The invention relates to a visual image set classification method. In particular, it relates to a transductive low-rank tensor discriminant analysis method considering finding a low-rank tensor discriminant representation on a Grassmannian manifold to improve the classification accuracy of an image set. Background technique [0002] In recent years, with the rapid development of camera technology and portable devices, a large number of image sets have emerged. An image set generally contains a certain number of images of the same thing, but due to different camera angles, different lighting conditions, or different noises, these Images will vary in appearance. Image set classification is a promising technique that has aroused great interest among researchers in the field of computer vision and has many applications, such as video supervision, action recognition, and face recognition. However, due to the existence of a large amount of redundant inform...

AI Technical Summary

Problems solved by technology

[0004] Although the above low-rank tensor methods have been successfully applied to many classification problems, they rarely add discriminative information, which has been proven to have a great effect on improving the accuracy of visual classification, so some people Some methods for low-rank tensor representation based on discriminative analysis have been proposed, e.g., Jia et al. [8] A Discriminative Low-Rank Tensor Representation for Action Classification and Image Restoration

However, these methods only consider the data representation of Euclidean space, and do not consider the essential manifold structure

Images