Research On Face Image Recognition Based On Pluse Coupled Neural Network

With the emergence of computer multimedia technology, human facerecognition is more and more get the favor of people. Pulse Coupled NeuralNetwork (Pulse Coupled Neural Network, PCNN) is closer to the real biologicalNeural networks, it has self supervision and self learning characteristics. Inrecent years, domestic and overseas scholars launched a series of research ofPCNN model and its application, the research result shows that the PCNNmodel has excellent result in feature extraction with rotation invariance andscale invariance, signal intensity invariance and other characteristics. In the fieldof image processing, PCNN showed excellent biological characteristics and itsapplication in the field of face recognition, has a unique advantage, therefore,research face image recognition problem based on the PCNN model hastheoretical significance and application value.In the process of face recognition, it is vulnerably influenced by externalconditions such as light intensity, the photographer postures or facialexpressions, these factors will have a significant impact on face recognitionresults. Face image feature extraction is a key step in face recognition, thereforethis paper propose PCNN pulse distribution of intensity (QD-PCNN) facefeature extraction method based on the simplified PCNN model, and use the gridsearch algorithm to optimize the parameters of QD-PCNN method, finallycombined with distance measurement function for face recognition. In this paper,the research work mainly includes the following aspects: 1．Analyze the problems existing in the traditional pulse coupled neuralnetwork，proposed a new method of feature extraction based on simplifiedPCNN model, extraction and analyze the feature of normalized face image usingQD-PCNN model, the results show that the extracted features can effectivelydistinguish the face image between different identity. Then calculate the averagefeature sequence of every class face image, and calculating the cosine distancebetween the average of all kinds of face image feature sequences and theunknown identify sequences, through the similarity judgment to realize facerecognition finally.2．Analyzes the intensity pulse distribution principle of PCNN model, nddiscuss the four parameters: weighting matrixW、link coefficient、thresholdattenuation and amplification、V in QD-PCNN network model respectively,then using the improved grid search parameters optimization method to find theoptimal combination of、 in QD-PCNN model, proved that the value ofparameters find by optimization method can achieve higher recognition accuracy,and it saves the computing time compared to the traditional grid search method.3．Designed and realized the PCNN face image recognition applicationplatform, the platform including the face image data acquisition module, PCAand ICA, PCNN face recognition module and PCNN parameters optimizationmodule. Platform realized the function of automatic detection and acquisitionface image, and analyzed the performance of the QD-PCNN feature extractionmethod proposed in this paper for the human face image recognition.