Research On Face Recognition Based On Pulse Coupled Neural Network

As an important aspect of biological feature recognition, face recognition is not only widely applied in public security, finance and so on, and has crucial theoretical research value. So how to extract facial feature effectively is a part of the hottest issues of face recognition technology. Pulse Coupled Neural Network (PCNN) has a biology background, which is hereby designated as the third generation of an artificial neural network. PCNN is a single network with self supervision and self learning, so it is not necessary for advanced learning. It shows the phenomena of pulse emission and dynamical threshold. All of these simulate well fatigue, refractory period and pulse stimulation to biology neural cells. More importantly, its feature sequences result in image feature extraction, and have the outstanding features of rotation invariance, scale invariance and so on. In this paper, after studying the Pulse Coupled Neural Network neural model, basic working principle and basic excellent characteristics, a method of face recognition based on PCNN feature sequences (time series, entropy sequences, logarithmic series, standard deviation series and mean residual sequences) is proposed.In this paper, we put five feature sequences of PCNN used in facial feature extraction and recognition. Using five feature sequences of PCNN to extract the feature sequences of each human face image and the feature sequences of every class are obtained through making averaging. Euclidean distance is expected to be adopted as a criterion for facial image recognition. From the experimental results, each kind of feature sequences carries much face information and description the face image very well. Combining with Euclidean distance, PCNN time series makes beat effects and gets the highest detection accuracy.After using the time series of PCNN to extra the feature sequences of each face image and recognize the face images in HSI color space. Color face image is transformed from RGB space to HSI space. Time series is extracted from each face image component in HSI space respectively. Then time series of each image component is connected which from the face image feature. Finally, Euclidean distance is expected to be adopted as a criterion for facial image recognition. Experimental result shows that this method is easy, and does not need for advanced learning. The algorithm has the biological visual characteristics, in line with the feeling and judgment of human visual system to the object, have the outstanding features of rotation invariance, scale invariance and so on, and the recognition method is not only simple but good recognition effect.