Hyung Lee  Sung Hyuk Choi  Jung Bae Kim  Jong Won Park

Vol. 8,  No. 1, pp. 9-15, Mar.  2001
10.3745/KIPSTA.2001.8.1.9

PDF

There have been studied on building various application systems related to image. However, low image processing speed in many systems previously developed has been the problem. We have investigated a parallel processing system for improving the processing speed and applied to many related applications. In this paper, we address parallel processing system designed as hardware for meeting the demand on image applications to be processed in real time. And, facial recognition system was selected and attempted using the proposed system in order to verify its performance. The parallel processing system is similar to a combined architecture of SIMD with MIMD. Since this combined architecture has flexibility and efficiency, it should be applied to many related image applications. The parallel processing system proposed consisted of 144 processing elements, and 12 multi-access memory systems, and two interface modules one is for an external processing unit, Intel960Kx processor, the other for external memory modules. The multi-access memory system we introduced is made up of a memory module selection module, a data routing module, and an address calculation and routing module. In order to verify the system, we developed a parallel algorithm for the facial image recognition using the mesh feature. The algorithm consists of a preprocessing to extract feature regions, normalizing the final feature region, extracting 4 feature vectors from the previous feature regions and the normalized region, and a classification. This algorithm was performed on the system. The system and all processing procedures of the algorithm on it were simulated and evaluated by the CADENCE Verilog- XL hardware simulation package.