A distributed parallel system for face recognition

We present a distributed parallel system for face recognition. The face database in this system is so huge (more than 1,000 thousand faces) that matching and recognition processes cannot be carried out on only one computer. So cluster system must be used to improve the matching speed. But many current clusters existing have some problems and cannot fit our system. A special distributed parallel system was developed to complete face query and recognition. The concept of parallel virtual machine and a kind of linked table structure adopted in this system not only decreased lots of moving overhead of adding or deleting nodes in the array structure but also truly realized the infinite extensibility. Furthermore, key techniques such as distributed database, buffer and synchronization techniques in communication and multithreading in control flow were adopted to guarantee the normal running. Practical results proved that this parallel system could improve the matching speed for more than 4 times. Moreover, the greatest advantage of this system is not only increasing matching speed but also breaking the upper limit of face data capacity. Consequently, the face data capability of this system can be extended to arbitrary figure as bigger as possible.