The Design And Implement Of A Hadoop Based Face Recognition System

Face detection and recognition is a hotspot in the research of modern biometrics recognition technology.With the development of identification and detection technology,a large number of commercial applications have emerged,such as face attendance and character search,which have greatly enrich people's production and life.With the popularity of mobile phones,computers and other intelligent devices,the traditional way of recording life by text has been gradually replaced by images,videos,etc.In face of such huge volume of data,how to find valuable things from these data is a problem to data management.In order to realize face detection and recognition under huge volume of data,we implement a face recognition system based on Hadoop.The system is divided into two parts: one is the parallel face detection based on Hadoop,and the other is parallel face recognition based on Hadoop.Face detection was implemented by using OpenCV and face recognition was implemented by adopting principal component analysis method.For face detection,we designed serialized image class because Hadoop did not provide the special image interface.Due to that a large amount of small files are not suitable for Hadoop to process,we realize the method of splite combination by overwriting recording read function which make small files into a logically big pieces.It effectively reduces the starting times of Hadoop function.This paper applied the OpenCV visual processing library into Hadoop and aims to realize face detection by using the trained classifier file.The experimental results show that the method of splite is helpful to reduce the overhead of Hadoop system and increase the speed of face detection.In the process of face recognition,according to the characteristics of Hadoop parallel computing framework,we select PCA as the method of feature extraction.Because a large number of mathematical operations were involved in feature extraction,we use the sequential file storage method to store the original matrix information and its characteristics which will meet the requirements of fast data reading.Finally,in order to increase the practicality of the program,we connect them through tomcat.