2016/04/25 by Mehrdad J. Gangeh, Safaa M. Bedawi, Gangeh, Mehrdad J. +5
Computer Science · Social Sciences · Engineering · #Face and Expression Recognition #Advanced Computing and Algorithms #Advanced Algorithms and Applications
paper · pdf · doi:10.48550/arxiv.1604.07319
In this paper, a novel semi-supervised dictionary learning and sparse\nrepresentation (SS-DLSR) is proposed. The proposed method benefits from the\nsupervisory information by learning the dictionary in a space where the\ndependency between the data and class labels is maximized. This maximization is\nperformed using Hilbert-Schmidt independence criterion (HSIC). On the other\nhand, the global distribution of the underlying manifolds were learned from the\nunlabeled data by minimizing the distances between the unlabeled data and the\ncorresponding nearest labeled data in the space of the dictionary learned. The\nproposed SS-DLSR algorithm has closed-form solutions for both the dictionary\nand sparse coefficients, and therefore does not have to learn the two\niteratively and alternately as is common in the literature of the DLSR. This\nmakes the solution for the proposed algorithm very fast. The experiments\nconfirm the improvement in classification performance on benchmark datasets by\nincluding the information from both labeled and unlabeled data, particularly\nwhen there are many unlabeled data.\n