DSP based modular face recognition system

In this study we develop a complete face recognition system using a CMOS camera and DM6437 Digital Video Development Platform. The face recognition system consists of several processing sub modules. First face region is detected in captured video frames. Then face regions are scaled into fixed sized images by using DSP VPFE resizer and are divided into overlapping sub-regions. Independent features are extracted by applying DCT to each sub region. Discriminating features are determined by Fisher Discrimating Value and variable length of features are selected from each region. Then faces are classified by Nearest Neighbour Classifier. Detected and identified faces are shown on the screen, connected to s-video output of the DM6437 EVM. Thus, by utilizing video processing environment of DM6437 we realize a low cost face recognition system.