项目名称: 基于表示学习的图像复原和识别方法研究
项目编号: No.61472187
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 其他
项目作者: 杨健
作者单位: 南京理工大学
项目金额: 87万元
中文摘要: 大数据中图像(视频)占据着举足轻重的地位,图像(视频)的理解和识别在社会、经济和国家安全等领域中扮演着越来越重要的角色。低质量和污染图像的识别已成为当前视觉监控和人脸识别领域的瓶颈问题。本项目将面向图像复原和识别两个紧密相关的问题,从低秩分解、稀疏表示和流形学习三方面开展表示学习理论与算法研究,旨在建立图像复原和分类识别一体化框架。主要内容包括:(1)基于核范数度量的最优重构方法研究;(2)基于表示学习的图像复原方法研究;(3)基于表示学习的特征生成方法研究;(4)图像复原和分类识别一体化方法研究。最后,基于以上理论与方法成果,构建图像复原和识别的稳健视觉系统验证平台。
中文关键词: 图像识别;特征提取;稀疏表示;流形学习
英文摘要: Due to the importance of image and video in big data, image and video understanding and recognition plan a critical role in our society, economy, and national security areas. Low-quality and contaminated images pose a challenging problem in video supervisory control and face recognition system. This project will be focused on the image recovery and recognition, the two closely related problems. We will do researches on low-rank decomposition, sparse representation, and manifold learning, aiming to build a unified framework for image recovery and recognition. The points of our research include: (1) nuclear norm based image reconstruction; (2) representation learning based image recovery; (3) representation learning based feature generation; (4) unified methods for image recovery and recognition. Finally, we will build a robust visual system to evaluate our theories and methods.
英文关键词: image recognition;feature extraction;sparse representation;manifold learning