Deep learning network for face detection

By the multi-layer nonlinear mapping and the semantic feature extraction of the deep learning, a deep learning network is proposed for face detection to overcome the challenge of detecting faces accurately and rapidly in the non-ideal case. Key to this deep network is that, to better simulate the response for information in the human brain, the status probability of the neuron is used to model the status of the human brain neuron which is a continuous distribution from the most active to the least active. Moreover, the number of the hidden layer´s neuron decreases layer-by-layer to eliminate the redundant information of the input data and accelerate the detection speed combining with the skin color detection. Experimental results show that, besides the fast detection speed and strong robustness to face rotation, the proposed method possesses lower false detection rate and lower missing detection rate.