Discriminative Dictionary Learning with Pairwise Constraints

Abstract

Abstract. In computer vision problems such as pair matching, only binary information -'same' or 'different' label for pairs of images -is given during training. This is in contrast to classification problems, where the category labels of training images are provided. We propose a unified discriminative dictionary learning approach for both pair matching and multiclass classification tasks. More specifically, we introduce a new discriminative term called 'pairwise sparse code error' for the discriminativeness in sparse representation of pairs of signals, and then combine it with the classification error for discriminativeness in classifier construction to form a unified objective function. The solution to the new objective function is achieved by employing the efficient feature-sign search algorithm. The learned dictionary encourages feature points from a similar pair (or the same class) to have similar sparse codes. We validate the effectiveness of our approach through a series of experiments on face verification and recognition problems.

Discriminative dictionary learning via shared latent structure for object recognition and activity recognition

“…Besides the difference in discriminative criteria, some methods trained dictionary learning and classifier parameters separately [13], [17], while others optimized dictionary and classifier parameters simultaneously [17], [23], [14], [25]. More recently, work by Guo et al [5] introduced a pairwise sparse code error into the discriminative dictionary learning framework. Despite the advances in incorporating discriminativeness in classifier construction, none of the aforementioned work addresses the tuning problem of the classifier weighting parameter or considers simultaneous dimensionality reduction for robust and efficient classification.…”

Learning a structured dictionary for video-based face recognition

“…Dictionary Learning: Dictionary learning [1,13,11,21,27,37,40,41,42,18] has attracted great interest in subspace modeling for classification purpose. It overcomes the limitation of PCA subspaces by using non-orthogonal atoms (columns) in the dictionary to provide more flexibility to model the data.…”

“…Recent works have shown that dictionary-based methods achieve impressive performance in various tasks, such as image-based face recognition, object and action recognition [1,13,11,21,27,37,40,41,42,18,35]. This is due to the fact that images could be well represented by an approximately learned dictionary and related sparse codes.…”

Face recognition via discriminative dictionary learning

“…However, SRC has a defect that its performance degrades for small dictionary size. To compensate the defect, several algorithms using discriminative dictionary learning and sparse coding have been proposed in [8,9,10,11,12,13]. Dictionary learning with sparse coding algorithms which were originally proposed for representing data with possibly few bases and sparse coefficients have been attracting considerable attention in recent years.…”