Sparse coding has been popularly used as an effective data representationmethod in various applications, such as computer vision, medical imaging andbioinformatics, etc. However, the conventional sparse coding algorithms and itsmanifold regularized variants (graph sparse coding and Laplacian sparsecoding), learn the codebook and codes in a unsupervised manner and neglect theclass information available in the training set. To address this problem, inthis paper we propose a novel discriminative sparse coding method based onmulti-manifold, by learning discriminative class-conditional codebooks andsparse codes from both data feature space and class labels. First, the entiretraining set is partitioned into multiple manifolds according to the classlabels. Then, we formulate the sparse coding as a manifold-manifold matchingproblem and learn class-conditional codebooks and codes to maximize themanifold margins of different classes. Lastly, we present a data point-manifoldmatching error based strategy to classify the unlabeled data point.Experimental results on somatic mutations identification and breast tumorsclassification in ultrasonic images tasks demonstrate the efficacy of theproposed data representation-classification approach.
展开▼
