Reconciliation of statistical and spatial sparsity for robust visual classification

Hao Cheng, Kim-Hui Yap, Bihan Wen

Abstract: Highlights•A novel joint sparse representation is constructed based on global statistical and local patch dictionaries from visual data.•The proposed method can be utilized for both image and image-set classification tasks with an efficient yet effective algorithm.•The proposed method achieves superior results on both image and imageset classification tasks.•Extensive experiments on few-shot and robust scenarios demonstrate the effectiveness and generalizibility of the proposed method.