在大数据中视觉信息（图像、视频）占据着举足轻重的地位，而视觉信息的局部特征表示又在图像识别/分类、图像检索和视频分析等领域扮演着重要角色。高效的视觉信息局部特征表示是提升图像（视频）分析系统性能的关键。本项目将重点研究视觉信息的局部特征表示问题，从局部结构表示、稀疏表示和稳健表示三个方面展开研究，旨在建立视觉信息局部特征表示的一体化框架。主要研究内容包括：（1）基于核范数的回归表示方法研究；（2）基于鉴别学习的局部特征表示方法研究；（3）基于稀疏稳健性的局部特征表示方法研究；（4）基于深层回归的局部特征表示方法研究。最后，基于以上理论和方法的成果，建立基于视觉信息局部特征表示的视觉系统验证平台。

Visual information (image and video) occupies an important position in big data. Local feature representation of visual information plays a central role in image recognition/classification, image retrieval and video analysis. It’s well known that how to design an effective local feature representation model for visual information is the key step to improve the abilities of image (or video) analysis system. This project will research on local feature representation models of visual information. Moreover, we will focus on provide an integrative framework of local feature representation according to the theories of local structure representation, sparse representation and robust representation. The main research content including: (1) nuclear norm based regression representation method; (2) discriminative learning based local feature representation method; (3) sparsity and robustness based local feature representation method; (4) deep regression based local feature representation method. Finally, we will build a vision system to verify our theories and methods.