基于低秩类间稀疏判别最小二乘回归的图像分类

在多分类任务中基于最小二乘回归(least squares regression,LSR)的分类器是有效的,但大多数现有方法因使用有限的投影而损失许多判别信息,有的算法只关注样本与目标矩阵的精确拟合而忽略了过拟合问题。为了解决这些问题并提高分类性能,本文提出了一种基于低秩类间稀疏性的判别最小二乘回归(low-rank inter-class sparsity discriminative least squares regression,LRICSDLSR)的多类图像的分类方法。在判别最小二乘回归模型中引入类间稀疏性约束,使得来自同一类的样本间隔大大减小,而来自不同类的样本的间隔增大;对由非负松弛矩阵获得的松弛标签施加低秩约束,以提高其类内紧凑性和相似性;在学习标签上引入了一个额外的正则化项,以避免过拟合问题。实验结果表明,这3个改进有助于学习明显的回归投影,从而实现更好的分类性能。

Image classification based on low-rank inter-class sparsity discriminant least squares regression

Classifiers based on Least Squares Regression(LSR)are effective in multi-classification tasks. However, most of the existing methods use limited projections and cause a lot of loss of discriminative information. Some algorithms only focus on the accurate fitting of the sample and the target matrix and ignore the problem of overfitting. In order to solve these problems and improve the classification performance, this paper proposes a multi-class image classification method based on low-rank inter-class sparsity discriminative least squares regression(LRICSDLSR). Introducing the inter-class sparsity constraint in the discriminative least squares regression model, so that the margin of samples from the same class can be greatly reduced, while the margin of samples from different classes can be increased; Apply low-rank constraints to the relaxed labels obtained from the non-negative relaxation matrix to improve their intra-class compactness and similarity; An additional regularization term is introduced on the learning label to avoid overfitting. Experimental results show that it helps to learn more distinguished regression projections to achieve better classification performance.The experimental results on a series of image data sets prove the effectiveness of the method.