【作者】 张明；

【导师】 刘祥楼；

【作者基本信息】 东北石油大学 ， 测试计量技术及仪器， 2011， 硕士

【摘要】 随着信息化技术、计算机技术、网络技术的迅速发展,可视化的人脸图像资源将变得越来越多。建立快速高效的人脸图像检索机制,己经成为电子信息领域内亟待解决的问题。本课题确定基于语义特征的人脸特征提取方法的研究,将运用数字图像处理技术、人脸模式识别技术与传统的数据库检索技术,融合基于文本的图像检索和基于内容的图像检索两者技术优势为一体,避开在图像之间进行的繁琐匹配处理过程。首先,采取基于结构的方法,用肤色自适应的方法从输入图像中自动找出人脸位置;其次,再利用原灰度图积分投影方法将面部主要器官从人脸图像中自动分割出来;然后,通过对面部器官的几何特征点检测来设定不同的面部器官特征参数,并按照这些特征参数的取值范围进行数据分类处理并转化为形象化的语义描述;最后,将这些形象化的语义描述组合在一起表示人脸,并设计一个语义人脸图像检索系统,通过语义数据库检索出一定量的相似人脸图像,从而实现快速准确的检索目的。以200人为待测样本进行检索实验,正确检索出来的为155人,完全匹配的为125人,系统的识别率为75.55%,平均检索时间<0.1s。实验证明,这种算法具有很强的鲁棒性,使用此方法所建立的语义人脸图像检索系统,具有快速、高效、实用的特点。更多还原

【Abstract】 With the rapid development of information technology,computer technology and network technology,visual face images resources will become more and more.Establishing a rapid and efficient mechanism for face image retrieval has become urgent problem in electronic information field.My paper determine the research of face features extraction methods based on semantic features.It will use digital image processing technology,facial recognition technology and traditional database retrieval techniques,fusing the advantages of text-based image retrieval and content-based image retrieval to avoid the tedious process of matching face image.First of all,structure-based methods,I use adaptive color mapping method from the input images automatically identify the location of face;secondly,use the original gray image integral projection method will pick up major face organs from the face images automatically;then the geometric characteristics of the face detection organ to set different facial organs parameters,in accordance with these range of parameters for data classification and translated into visualization of the semantic description;Finally,the visualization of the semantic description compose a face together and design a semantic face image retrieval system which can fast search through semantic database to retrieve a certain amount of similar face images. In order to achieve fast and accurate retrieval purposes.With 200 human being tested samples retrieval experiment,correct retrieved is 155,exactly match is 125,system recognition rate is 75.55%,average retrieval time < 0.1s. Experimental results show that the algorithm has strong robustness,use this method to establish the semantic face image retrieval system with fast,efficient and practical character.更多还原