Joint feature selection and graph regularization for modality-dependent cross-modal retrieval.

•We simultaneously consider the important differences of sub-retrieval tasks and the discriminative semantics latently involved in multi-modal features when learning the shared subspace for cross-modal retrieval.•We seamlessly integrate the linear regression term, correlation analysis term, graph regularization term and feature selection term into a joint cross-modal learning framework.•An iterative algorithm guaranteed with convergence is proposed to solve the formulated optimization problem.