【计算机科学】【2017.01】基于深度学习的人脸识别

本文为西班牙加泰罗尼亚理工大学（作者：Xavier Serra）的硕士论文，共96页。

人脸识别是目前发展起来的具有多种现实应用的技术。本硕士论文的目标是为一家人工智能公司GoldenSpearLLC开发一个完整的人脸识别系统。该系统利用卷积神经网络提取人脸相关特征，这些特性允许以有效的方式比较他们的面部。该系统可以通过整合已经处理的新人员，以改进它对已有人员的预测，实现了在线识别与在线学习同时进行。在一组100人的测试中，准确率已经超过了95%，并且已经被证明可以随着系统中人数的增加而提高准确率。我们提供了两个基于人脸识别的应用程序。

Face Recognition is a currently developingtechnology with multiple reallife applications. The goal of this Master Thesisis to develop a complete Face Recognition system for GoldenSpear LLC, an AIbased company. The developed system uses Convolutional Neural Networks in orderto extract relevant facial features. These features allow to compare facesbetween them in an efficient way. The system can be trained to recognize a setof people, and to learn in an on-line way, by integrating the new people itprocesses and improving its predictions on the ones it already has. Theaccuracy in a set of 100 people has surpassed the 95%, and it has proven to robustlyscale along with the number of people in the system. We provide twoapplications we have developed that make use of this Face Recognitiontechnology.

1. 引言
2. 人脸识别问题
3. CNN的理论背景
4. 我们的设计方案
5. 实验与结果
6. 结论
   附录A 数据集名称列表

更多精彩文章请关注公众号：