基于Hadoop平台的多模态人脸识别研究

Research on Multimodal Face Recognition Based on Hadoop Platform

摘要: 随着现代人越来越喜欢用图像记录生活，每日上传至互联网的图像都呈爆炸式增长。公安部门可以利用海量的图像数据实现网络追凶，但现实中摄像头采集到的图像以及上传至网络的图像，并非都是统一状态的图像，而是包含各种状态的图像，例如不同表情、不同动作、不同角度、不同程度的角度偏斜，不同年龄，以及有背景干扰的图像，其中部分图像会因使用的设备不同，导致上传至网络的图像既有灰白图像又有彩色图像。这些多模态图像为人脸的准确识别增加了难度，要将实际中采集到的图像从如此复杂且规模庞大的数据集中匹配与识别出来，是一件十分困难的事。针对以上问题，提出将改进后的开源人脸识别库，即face_recognition库与Hadoop平台中的MapReduce进行结合，在确保识别准确率的前提下提升人脸检测速度，实现对大规模、多模态图像的有效识别。实验证明，本文的方法能够有效解决大规模多模态图像的识别问题，实时性高，实用性强。

Abstract: The number of images uploaded to the Internet every day is exploding as modern people increasingly document their lives in images. Public security department can take advantage of the huge amounts of image data to realize the network after fierce, but in reality, the camera collected images and images uploaded to the Internet, are not all images of the united state, but contain images of the various states, such as different expressions and actions, different angles, different degree of skew Angle, different age, as well as the background interference image. Some of the images are uploaded to the network due to different devices, resulting in both gray and color images. These multi-modal images increase the difficulty of accurate face recognition, and it is very difficult to match and recognize the images collected in practice from such a complex and large data set. To solve the above problems, this paper proposes to combine the improved open source face_recognition library, namely face_recognition library, with MapReduce in Hadoop platform to improve face detection speed and realize effective recognition of large-scale and multi-modal images on the premise of ensuring recognition accuracy. Experimental results show that the proposed method can effectively solve the problem of large-scale multi-modal image recognition, with high real-time performance and practicability.