人脸定位及分合算法研究.docVIP

基于几何匹配和分合算法的人脸识别 目前在人脸定位领域，几何模型匹配方法得到了广泛的使用。 本文对几何模型匹配方法进行了研究，提出了一套完整的人脸定位算法。在预处理部分，采用了特别的增强人脸特征与脸部皮肤之间对比度的方法及局域取阈值二值化方法，改进了预处理的效果。在图像分割部分，实现了经典的分合算法，并且使用成组算法改进了分合的效果。在人脸匹配部分，实现了基于眼睛和嘴的几何模型匹配，并对评价函数的构造进行了研究。 在Microsoft Windows 98平台上Borland Delphi 3.0开发环境下，采用上述算法编制了人脸定位的软件。实验证明，该软件对于尺寸足够大的清晰的正面单个/多个人脸情形能够正确检测定位，典型耗时为3～40秒钟。 最后，本文探讨了几何匹配算法和分合算法的优缺点，提出了进一步工作的方向。 关键词：人脸定位，几何匹配，分合算法，二值化，图像分割，成组算法，评价函数。 Abstract In the area of human face locating, methods using a geometric face model(geometrical matching) are widely used. This paper studied a method of geometrical matching and presented a complete set of algorithms for human face locating. In the section of preprocessing, special methods to enhance the contrast between face features and face skin and a local thresholding method were employed so that the result of the preprocessing was improved. In the section of image segmentation, the classical split-merge algorithm was applied, and the result improved by a grouping algorithm. In the section of face matching, an algorithm using a geometric model based on eyes and mouth was presented. Discussion was conducted on the construction of evaluation functions. Under Microsoft Windows 98 and Borland Delphi 3.0, a face locating software was developed using the algorithms mentioned above. Experiments showed that the software is capable of detect and locate front-view face(s) which is(are) big and clear enough. The typical time consumed is 3～30 seconds. In the end, this paper probed into the advantages and disadvantages of the geometrical matching and the split-merge algorithms, and put forward suggestions to future work. Key words: Face locating, geometrical matching, split-merge, thresholding, image segmentation, grouping, evaluation function. 基于几何匹配和分合算法的人脸识别 摘 要 1 Abstract 2 目 录 3 第一章 引言 4 1.1 人脸定位概述及其与人脸识别的关系 4 1.2 人脸定位算法 5 1.3 分割算法和区域生长 6 1.4 论文的主要工作 7 第二章 一种有效的分割算法: 分合算法 8 2.1 人脸定位中的图像分割 8 2.2 分合算法描述 9 2.3 成组算法 13 2.4 形状属性的计算 14 第三章 预处理和几何匹配 19 3.1 预处理 19 3.2 几何匹配算