项目名称: 面向图像与视频特征表示的深度编码方法研究
项目编号: No.61272319
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 常虹
作者单位: 中国科学院计算技术研究所
项目金额: 80万元
中文摘要: 本项目拟在深度学习、稀疏编码、以及图像与视频表示的研究成果的基础上,针对图像与视频特征所具有的高维、非线性、问题依赖等难题,研究新的深度编码的理论和方法,以及在具体的应用问题中所需的特定技术。具体而言,我们的研究目标是(1)提出新的深度编码框架和算法,并对算法的各项性能做出分析与证明;(2)提高深度编码方法的可扩展性,降低其计算复杂度;(3)面向特定的应用问题,有针对性的提出基于深度编码的图像和视频特征表示方法,提高监控视频中的人体检测与检索、动作识别等应用系统的识别精度、泛化能力以及鲁棒性能。本项目发表期望产出高水平论文和有效的应用系统。
中文关键词: 图像视频表示;深度学习;流形学习;计算机视觉;
英文摘要: Based on the recent related work on deep learing, sparse coding, image and video representation, this project proposes to research on deep coding theory and algorithms as well as their application in real-world recognition problems, to deal with the challenges of high-dimensionality, nonlinear, and problem dependent in image and video feature representation. More specifically, we aim to (1) propose novel deep coding framewpork and approaches, analyze and verify their properties; (2) improve the scalability and computational efficiency of the proposed deep coding methos; (3) propose deep coding based image and video feature representation methods for specific application probems, in order to improve the recognition accuracy and generalization ability of real-world recognition systems, such as human detection and retrieval in video surveillance, action recognition, and so on. This project aims to output high quality research papers as well as effective real-world application systems.
英文关键词: image video representation;deep learning;manifold learning;computer vision;