项目名称: 基于稀疏视觉特征学习的真实环境人脸识别研究
项目编号: No.61203247
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 自动化学科
项目作者: 赵才荣
作者单位: 同济大学
项目金额: 24万元
中文摘要: 随着数码产品、智能手机的普及,人脸图像数据越来越容易获得,如何利用好这些图像信息有着迫切的市场需求,真实环境人脸识别技术是解决该需求的有效方法之一。 真实环境人脸识别技术研究的难点和重点之一是如何抽取一种对真实环境人脸自身变化和摄影环境变化鲁棒的视觉特征。本项目以建立基于真实环境人脸的视觉感知计算模型、提取真实环境人脸的显著特征和Gist特征和稀疏视觉特征学习为主要攻关内容,借鉴视觉感知领域的研究成果,对真实环境人脸识别问题进行应用基础研究探讨。 本项目的创新之处是:(1)设计一种基于粒计算思想的初级视觉筛选规则,构建一种新的适合真实环境人脸的视觉感知计算模型;(2)提出真实环境人脸的显著特征和Gist特征概念并给出相应特征抽取算法;(3)提出一种稀疏视觉特征学习算法进行鉴别分析。本项目的开展将为真实环境人脸识别奠定理论和技术基础,并为其他系统对复杂环境的自适应能力提高提供新的参考方法。
中文关键词: 稀疏学习;视觉特征;图嵌入;不相关鉴别分析;
英文摘要: With the widely uses of digital products and smart phones,the real-world face images become more and more easier to obtain. Thus, there has been an emerging market demand on how to use these images. One of the effective techniques to solve this problem is the robust face recognition algorithms that are able to deal with real-world face images. One of the difficult but important points in the real-world face recognition research is on how to extract the visual features robust to the facial and circustant variations.Based on the results in visual perception, this research discusses the real-world face recognition problems encountered in basic applications and aims to build the sparse visual perception model for real-world face recognition, to extract the salient features and gist features, and to learn the sparse visual features. The creavities of this research are as follows: (1) design a selection criterion on early visual features based on granular computing and build a visual perception computing model to fit the real-world conditions encountered in face recognition;(2)propose the concepts on the salient features and the gist features of the real-world face images and present the algorithms on how to extract these features; (3) propose a novel sparse visual feature learning algorithm for discrimination. The pu
英文关键词: Sparse learning;visual feature;graph embedding;uncorrelated discriminant ananlysis;