Locality-constrained affine subspace coding for image classification and retrieval

•We propose a novel feature coding method based on a dictionary of affine subspaces in terms of feature reconstruction and statistics modeling.•The proposed method models the first- and second-order information, while preserving the properties of locality and sparsity of coding vectors.•Extensive experiments show the proposed method performs on par with or better than the state-of-the-arts for image classification and retrieval.