• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Conference on Artificial Intelligence IC-AI'2000 Vol.3, Jun 26-29, 2000, Las Vegas, Nevada, USA >Benchmark testing and performance improving using neural network for face recognition
【24h】

Benchmark testing and performance improving using neural network for face recognition

机译：使用神经网络进行人脸识别的基准测试和性能改进

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

The benchmark test result for face recognition using PCA (principal component analysis) with Euclidean distance shows that the recognition result is unable to meet certain accuracy. As the result of PCA with Euclidean distance shows high EER (equal error rate), we propose a neural network using recognition by recall approach to improve the performance. We give a training method of neural network to tolerant the eye position error. The result shows that the verification performance of the neural network using recognition by recall is much better that PCA with Euclidean distance. The benchmark testing is based on two sets of very different types of images: University of Surrey images with 295 persons, each person has 4 different photos. Each of them is taken with one month difference with similar lighting condition and slightly different facial expression and orientation. The other images set (here we call it BDIS - Big difference image set) contain 1955 different persons that each person has two photos which are taken with 10-20 years time difference. We give the FAR, FRR and EER for the two data sets based on PCA with Euclidean distance measure and the dramatically improved result with neural network using recognition by recall approach. We also show the eye location effect and image time effect based on the University of Surrey images.
机译：使用具有欧几里德距离的PCA（主成分分析）进行人脸识别的基准测试结果表明，该识别结果无法满足一定的准确性。由于具有欧氏距离的PCA的结果显示出较高的EER（均等错误率），因此我们提出了一种利用召回识别方法来改善性能的神经网络。我们给出了一种神经网络的容忍眼位误差的训练方法。结果表明，使用召回识别的神经网络验证性能要优于具有欧氏距离的PCA。基准测试基于两组非常不同类型的图像：萨里大学的295人图像，每人有4张不同的照片。他们每个人的采光条件相似，面部表情和朝向也相差一个月。其他图像集（这里称为BDIS-大差异图像集）包含1955个不同的人，每个人有两张以10到20年的时差拍摄的照片。我们给出了基于PCA和欧氏距离测度的两个数据集的FAR，FRR和EER，以及使用召回方法识别的神经网络显着改善的结果。我们还根据萨里大学的图像显示了眼睛的位置效应和图像时间效应。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号