基于神经网络和傅里叶描述子的残缺面部图像识别与分析  

Recognition and analysis of incomplete facial images based on neural network and Fourier descriptor

作　　者：吴卓 梁珂 张春阳 Wu Zhuo;Liang Ke;Zhang Chunyang(Beijing Information Science and Technology University,Beijing 100000,China)

机构地区：[1]北京信息科技大学,北京100000

出　　处：《无线互联科技》2021年第1期53-59,共7页Wireless Internet Technology

基　　金：北京信息科技大学2020年大学生创新创业训练计划项目,项目编号:5102010805。

摘　　要：传统的人脸识别系统中,往往都是使用完整图像进行对比和查找。但是在医疗或者刑侦领域,有时会出现残缺的人脸正面图像无法与已存入系统中的完整人脸图片相对应的情况。文章在传统的傅里叶描述子的基础上,提出相关改进算法。根据传统的傅里叶描述子提取特征简单的优点,改进传统的傅里叶描述子,同时进一步结合多任务卷积神经网络与图像分割的相关算法,进一步分析图像之间的关联关系和图像之间的相似度。实验表明,在已经收集到图片数据集中使用此算法,准确率可以达到90%~95%。识别率较之传统方法,得到了很大的提升。In traditional face recognition systems,complete images are often used for comparison and search.However,in the field of medical treatment or criminal investigation,sometimes the mutilated front face image cannot correspond to the complete face image stored in the system.On the basis of the traditional Fourier descriptor,related improved algorithms are proposed.According to the advantages of simple feature extraction of traditional Fourier descriptors,the traditional Fourier descriptors are improved,and the multitask convolutional neural network and related algorithms of image segmentation are further combined to further analyze the relationship between images and the relationship between images.The similarity between.Experiments show that using this algorithm in the collected image data set,the accuracy can reach between 90%and 95%.Compared with traditional methods,the recognition rate has been greatly improved.

关 键 词：傅里叶描述子 五官识别 人脸分割 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]