人脸识别是一种根据人类的面部特征来鉴别个体的技术.其中需要通过摄像机获取人脸图像,但在产生大量的小文件过程中,过去的分布式文件系统很难为其提供高性能读写和快速检索.结合FastDFS,Redis以及Mysql来优化图片的存储和索引,将同一摄像机目录下的小图片文件合并成大文件,并在其中建立内部小文件索引,然后将合成的大文件写到FastDFS中生成大文件索引,最后客户端结合小文件索引和大文件索引生成全文索引,并且利用Mysql的持久性存储特点来进行所有文件名和对应全文索引的存储以及利用Redis内存数据库来暂存近一年的文件和读取文件,同时采取提前读的机制来提前预取相邻时间的文件存放到客户端缓存来减少IO的操作.最后通过实验证明写性能平均提高了7.5％,读性能平均提高了5.0％.%Face recognition is a technique to identify individuals based on their facial features. The face images are captured by the camera mainly. However, in the process of generating a large number of small files, it is difficult for the distributed file system to provide high performance reading and writing and fast retrieval. In this paper, combining FastDFS, Redis and Mysql to optimize the storage and index of images, small image files in the same camera directory are merged into large files, and internal small file index is established. The resultant large file is then written to FastDFS to generate a large file index. At last, the client combines the small file index with the large file index for generating the full-text index, taking advantage of the Mysql persistent storage characteristics for the storage of file name and corresponding full-text index, using the Redis memory database to temporarily store files and read files for nearly a year, and at the same time, adopting the pre-reading mechanism to pre-fetch the peripheral time files to reduce the operation of IO. Finally, it is proved that the average writing performance is improved by 7.5％, and the average reading performance by 5.0％.
展开▼
