Implementation of face selective attention model on an embedded system

Abstract

This paper proposes a new embedded system which can selectively detect human faces with fast speed. The embedded system is developed by using OMAP 3530 application processor which has DSP and ARM core. Since the embedded system has the limited performance of CPU and memory, we propose a hybrid system combined the YCbCr based bottom-up selective attention with the conventional Adaboost algorithm. The proposed method using the bottom-up selective attention model can reduce not only the false positive error ratio of the Adaboost based face detection algorithm but also the time complexity by finding the candidate regions of the foreground and reducing the regions of interest (ROI) in the image. The experimental results show that the implemented embedded system can successfully work for localizing human faces in real time. © 2012 Springer-Verlag.