Research And Implementation Of Methods For Parallel Processing Of Face Image Recognition Based On Hadoop

In recent years, network image data surges with the development of network. How to effectively analyze and process the vast amount of image data has become a hot research topic in many fields. As a key component of massive image data, face image is an important basis for various surveillance and security fields to identify the individual. When facing huge face database, how efficient face recognition has become a critical issue in the surveillance and security field. Therefore, research on the massive efficiency face recognition algorithms is a significant theoretical and practical topic.Cloud computing currently is the cutting-edge technology of processing massive data. Hadoop is its representative, which has been widely used to process massive big files in various fields and achieved good results. However, research on massive small file processing (such as face recognition) is not mature. In this thesis, Hadoop technology is introduced to study efficiency algorithms for massive face recognition. Performance and advantages of cloud computing on massive small file processing are analyzed, and its optimization strategy is put forward which may help to extend the application areas of Hadoop. The main contents of this thesis are as follows.This thesis firstly carries on the selection of facial feature extraction methods. Face recognition performance evaluation is operated on four common face databases, e.g., FERET, ORL, Yale, AR based on typical feature extraction methods. Local direction pattern with good performance is selected as the method in the whole face recognition process by experiments, which lays the foundation for the following work. Secondly, how to use Hadoop to parallel process face data is investigated. Through in-depth study of the Hadoop framework, it is found that it does not provide an IO interface for image processing. When directly dealing with massive small files, Hadoop consumes much time and memory. To solve these problems, the combination file split technology provided by Hadoop is introduced in the process of small files. Through a combination of small files into large files in a logical way, the problem of low efficiency of Hadoop when reading the massive small files is solved. The feasibility and efficiency of the proposed approach are verified by parallel experiments. Further studies show that the method has the drawback of high memory consumption. SequenceFile method is then introduced. The massive small files are packaged in the form of large files to storage and process by Hadoop. The advantages of Hadoop for processing big data are fully used, which reduces the store pressure of massive small files brought to HDFS. Experimental results show that the scheme can easily deal with massive small pictures in Hadoop. Finally, face recognition parallelization is achieved based on Hadoop. Experimental results validate that the use of Hadoop may enhance the efficiency of the entire process of face recognition which provides a feasible solution for large-scale applications of face recognition.