当前位置:首页>> >>


基于混合特征图像处理的蛾类识别.rar

收藏

资源目录
    文档预览:
    编号:20181030002154701    类型:共享资源    大小:66.87MB    格式:RAR    上传时间:2018-10-30
    尺寸:148x200像素    分辨率:72dpi   颜色:RGB    工具:   
    25
    金币
    关 键 词:
    基于 混合 特征 图像 处理 识别
    资源描述:
    摘要I摘要人脸识别作为生物特征识别的一个重要分支,一直是模式识别和图像处理领域的研究热点之一。本文主要研究了基于几何特征的三维人脸识别方法。首先通过双目立体视觉技术获取了具有良好表达能力的人脸点云数据。在此基础上,通过准确提取人脸轮廓线和定位特征点,得到了表征人脸的三维几何特征。最后利用得到的几何特征作为三维人脸识别的依据。本文的主要研究内容和工作总结如下:(1) 点云数据获取:针对人脸这一特殊的立体匹配场景,研究了多种立体匹配方法,并做了分析和比较。运用改进的秩变换方法和金字塔匹配模型获取了高分辨率人脸致密视差图。结合前期摄像机标定后得到的摄像机参数,重建人脸三维点云数据,并对原始点云数据进行规格化处理,最后得到了具有良好表征能力的点云数据,并创建了三维人脸数据库 3DFACE-XMU。(2) 特征点定位和几何特征计算:利用深度信息准确提取人脸中分轮廓线和鼻尖横切轮廓线,并对轮廓线进行曲率计算和分析,结合人脸先验知识,定位出了人脸中分轮廓线上的十个特征点和鼻尖横切轮廓线上的三个特征点。针对鼻子等人脸的刚性区域,选取并计算了 4 类(包括曲率、距离、体积和角度)共 13 维的特征向量,作为后期三维人脸识别的依据。(3) 三维人脸识别系统的构建:基于 Visual C++ 、MATLAB 和 SQL SERVER平台开发了三维人脸识别系统,包括点云数据获取模块和识别模块。在 ZJU-3DFED 和 3DFACE-XMU 开展了基于几何特征的三维人脸识别实验,在数据较好的ZJU-3DFED 数据库上(其中训练样本 142 个,测试样本 142 个),rank-1 取得了 87%的识别率。关键词:双目立体视觉;点云数据获取;三维几何特征;人脸识别AbstractIIIAbstractAs an important branch of biological feature recognition, face recognition has been one of the most popular topics in the field of pattern recognition and image processing. The research on 3D face recognition in this paper is mainly based on geometric features. We first obtain the 3D face point cloud with good expression ability by binocular stereo vision. On the basis of this step, 3D geometric features are acquired by extracting face profiles and locating face feature points. Finally, we carry out a 3D face recognition method based on these geometric features. The main contributions of our work are as follows.(1) 3D face point cloud acquisition: Considering the fact that face is a special stereo matching sence, we research three stereo matching methods and make comparison and analysis of these methods. An improved rank transform based method and pyramid matching model are applied to gain the dense high resolution disparity map. Combining with the camera parameters achieved by the previous camera calibration step, we reconstruct the original point cloud. Then, by normalizing the original 3D face point cloud, we get much less amount of points, which still contain the main characteristics of the face. Finally, a database named 3DFACE-XMU is created.(2) feature points location and geometric features calculation: The central vertical profile and the nasal tip transverse profile are extracted accurately by the depth information, and then ten feature points on the central vertical profile and three feature points on the nasal tip transverse profile are located based on the prior knowledge and the curvature value. For the rigid region of face such as nose, we have selected and calculated four kinds of geometric features, totally 13-dimensional feature vector, including curvature, distance, volume, area and angle, which is used for 3D face recognition later.(3) Develop a 3D face recognition system: Based on the technology of Visual C++, MATALB and SQL SERVER, we have developed a 3D face recognition system including point cloud acquisition module and recognition module. The experiments AbstractIVare carried out in ZJU-3DFED and 3DFACE-XMU. In the better database ZJU-3DFED, which includes 142 training samples and 142 testing samples, rank one has 87 percent accuracy.Keywords: binocular stereo vision; point cloud acquisition; 3D geometric feature; face recognition
    展开阅读全文
    1
      金牌文库所有资源均是用户自行上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作他用。
    0条评论

    还可以输入200字符

    暂无评论,赶快抢占沙发吧。

    关于本文
    本文标题:基于混合特征图像处理的蛾类识别.rar
    链接地址:http://www.gold-doc.com/p-254801.html
    关于我们 - 网站声明 - 网站地图 - 资源地图 - 友情链接 - 网站客服客服 - 联系我们
    [email protected] 2014-2018 金牌文库网站版权所有
    经营许可证编号:浙ICP备15046084号-3
    收起
    展开