Transfer Subspace Learning based on Double Relaxed Regression for Image Classification

doi:10.1007/s10489-022-03213-z

Transfer Subspace Learning based on Double Relaxed Regression for Image Classification

2022-03-22

8

PRE

AI

DOI：10.1007/s10489-022-03213-z

Original

Original request for help

Save

Abstract

En 中文

A novel method based on relaxed regression and transfer subspace learning for cross-resolution image classification is presented. Firstly, a transfer subspace learning based on the double relaxed regression (TSL_DRR) method is adopted to learn a discriminative model and simultaneously avoid over-fitting in a regression-based classification task. Secondly, the matching efficiency between low-resolution face and high-resolution face is not ideal, so a so-called transfer subspace learning (TSL) technique is introduced to the proposed method to ensure that the domain data can be better matched by projecting different resolution face images onto the common subspace. Lastly, the global data structure and local data structure can be reliably retained by applying the low-rank and sparse constraint matrices, which also reduces the noise to an extent. Extensive experiments on various real image data sets indicate that the proposed method is effective in 4.2 accuracy.

Keywords:

Low-rank and sparse constraints

Transfer learning

Image classification

Cross-resolution

Transfer Subspace Learning based on Double Relaxed Regression for Image Classification

Abstract

Journal

Organization