A novel parallel image processing method on multi-core processors and its application on face recognition problem

Increasing processing speeds and number of processors' cores open a new door to run programs that require much processing in real-time. However, the increase in core numbers aims multitasking rather than performance improvement. In this thesis, a novel parallel image processing and frame rate stabilization approach for processing images in real-time is proposed. The main goal of this approach is to run real time image processing tasks on a regular PC with a multi-core CPU without requiring explicit hardware support. Various face detection and recognition algorithms are studied and implemented to run in parallel, using the presented approach. The approach is implemented under Microsoft .NET Framework and tested on Microsoft Windows 7 operating system with an Intel i7-3770 3,40 GHz CPU. Results show that the image processing tasks were successfully distributed to the cores and the performance is increased up to 5,25 times on 4 physical cores. The algorithm is also tested on face recognition task. It was shown that, workloads were successfully distributed to the cores and desired output frame rates were obtained for 25 and 60 fps videos.