基于人脸认知模式的相似脸搜索

发布时间：2018-10-24 10:35

【摘要】：人类对人脸认知模式的探索由来已久,并且已经成功应用于美容整形等研究领域。而计算机视觉和模式识别领域对人脸相似度度量方法的探索也从未停止,由此产生的人脸识别、人脸检索、相似脸搜索等应用络绎不绝。但是,现有的人脸相似度度量方法没有考虑人对人脸的认知模式,使得现有方法的计算结果从人的认知习惯角度来讲并非最佳。本文通过分析人脸认知模式,将各面部器官(眼睛、鼻子、嘴巴、脸型)的常见类型定义为标准型(例如,眼睛的标准型包括丹凤眼、吊眼等),为面部各器官构建标准型空间,将人脸图像的各个面部器官分别映射为相应的标准型空间中的一个点,依据两张人脸图像中各面部器官在标准型空间中的点坐标度量其形状相似度；采用圆形LBP算子,计算两张人脸对应器官的局部纹理相似度；二者综合作为相似脸搜索的依据。分别用本文方法和代表相似脸搜索最高水平的Face++的方法对80张正面、中性表情、平视角度拍摄的人脸图像进行测试,本文方法的TOP1,TOP2最相似搜索结果的准确率均明显高于Face++方法。实验结果表明,本文方法的搜索结果更加符合人脸认知模式。此外,还可以将此类基于认知模式的图像搜索思路推广应用于商业领域,如基于图像的相似网购商品搜索等。
[Abstract]:Human face recognition model has been explored for a long time, and has been successfully applied to cosmetic plastic surgery and other research fields. In the field of computer vision and pattern recognition, the exploration of face similarity measurement has never stopped, resulting in an endless stream of applications such as face recognition, face retrieval, similar face search and so on. However, the existing methods of face similarity measurement do not take into account the cognitive pattern of human face, which makes the calculation results of the existing methods not the best from the perspective of human cognitive habits. In this paper, the common types of facial organs (eyes, nose, mouth, face) are defined as standard types (for example, the standard types of eyes include Danfeng eye, hanging eye, etc.). Each facial organ of the face image is mapped into a point in the corresponding standard space, and the shape similarity is measured according to the point coordinates of each facial organ in the standard space in the two face images, and the circular LBP operator is used to measure the shape similarity of the face organ. The local texture similarity of the corresponding organs of two faces is calculated, and they are used as the basis for searching similar faces. Using the method of this paper and the method of searching for the highest level of Face on behalf of similar faces, we tested 80 face images taken from frontal, neutral and level angle. The accuracy of most similar TOP1,TOP2 search results of this method is obviously higher than that of Face method. Experimental results show that the search results of this method are more in line with the face cognitive mode. In addition, this kind of image search method based on cognitive mode can also be extended to the commercial field, such as similar online shopping based on image search and so on.
【学位授予单位】：北京林业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】