Research On Human Face Detection Algorithm Based On Hybrid Features And Implementation Using FPGA

With the development of human face detection algorithms, the existing face detection technology has great achievements in the field of software. It has been widely used in human-computer interaction. The effect of frontal face detection has reached to a practical level, however, the effect is limited in terms of multi-angel and multi-pose human face detection. Based on Viola and Rainer’s work, this paper investigates the method of improving the face detection rate. The hybrid features which combines extended Haar-like features and skin color features is introduced. And implemented this algorithms on FPGA. This paper mainly includes the following aspects:Firstly, the extended Haar-like features is trained by using the multi-angle and multi-pose human face library. After that, the classifier has practical detection rate with multi-angle and multi-pose human face library. However, in this way, its false positive rate also been increased.Secondly, in order to lower the false positive rate which is caused by the extended Haar-like features, Skin color features are brought into the algorithm. The algorithm is tested in PC simulation. Result shows that the face detection algorithm which is based on hybrid features has ideal detection rate and false positive rate in multi-angle and multi-pose human face library.Finally, based on hybrid features, this paper proposes design and implementation of the human face detection classifier FPGA system. This design is programed by Verilog and verified by Altera’s Cyclone Ⅱ FPGA. After serial communication and FPGA development board combined test, the data shows implementation which is proposed in this paper performs well. For detection rate and false positive rate is ideal, the test result is consistent with the theoretical algorithm.The PC simulation results and FPGA implementation show that the proposed approaches provide better detection rate compared to conventional approaches. FPGA implementation also achieve the prospective effect.