PSSDL: Probabilistic Semi-supervised Dictionary Learning

Abstract

Abstract. While recent supervised dictionary learning methods have attained promising results on the classification tasks, their performance depends on the availability of the large labeled datasets. However, in many real world applications, accessing to sufficient labeled data may be expensive and/or time consuming, but its relatively easy to acquire a large amount of unlabeled data. In this paper, we propose a probabilistic framework for discriminative dictionary learning which uses both the labeled and unlabeled data. Experimental results demonstrate that the performance of the proposed method is significantly better than the state of the art dictionary based classification methods.

Semi-supervised dictionary learning with label propagation for image classification

“…Zhang et al [22] proposed an online semi-supervised dictionary learning framework which integrated the reconstruction error of both labeled and unlabeled data, label consistency, and the classification error into an objective function. Babagholami-Mohamadabadi et al [23] integrated dictionary learning and classifier training into an objective function, and preserved the geometrical structure of both labeled and unlabeled data. Recently, Wang et al [21] utilized the structural sparse relationships between both the labeled and unlabeled samples to learn a discriminative dictionary in which the unlabeled samples are automatically grouped into different labeled samples.…”

“…Here, as in prevailing semi-supervised dictionary methods [11,18,19,[21][22][23]36], we assume that the unlabeled training data belongs to some class of the training set. In our proposed model, the dictionary to be…”

“…Dictionary learning methods can be divided into three main categories: unsupervised [13], supervised [14][15][16][17], and semi-supervised [11,[18][19][20][21][22][23]. K-SVD [13] is a representative unsupervised dictionary learning model, which is widely applied to image restoration tasks.…”

“…Semi-supervised dictionary learning [11,18,19,[21][22][23]36] has gained considerable interest in the past several years. In semi-supervised dictionary learning, the issue of whether the unlabeled samples can be accurately estimated for use as labeled samples for training is very important.…”

“…In Ref. [23], the unlabeled samples are utilized to learn a discriminative dictionary by preserving the geometrical structure of both the labeled and unlabeled data. However, the class-specific reconstruction error which carries strong discriminative ability cannot be utilized to estimate the identities of unlabeled samples in the shared dictionary.…”