Research On Face Recognition Of Infrared Image

Face recognition of infrared image is a technology of identification and classification oninfrared face image which collected by thermal infrared imaging device by using the methodsof pattern recognition, image processing, etc. The technology of infrared image facerecognition is a extension to the visible image face recognition. The fusion of both facerecognitions will be an important direction of face recognition. Based on the analysis ofcharacteristics of infrared face image, several methods of infrared image face recognition arestudied systematically in this paper.The preprocessing of infrared face image is researched. Eye's position is located byhands. Geometric preprocessing including rotation, clipping and scaling, grey valuepreprocessing including histogram equalization are adopted.Two methods of infrared image face recognition based on Principal Component Analysisand Linear Discriminant Analysis is studied. Dimensionality of the infrared face images arereduced, and the feature vectors are extracted, the feature vectors are used to generating aclassifier which can minimize within-class scatter and maximize between-class scatter.Experiments demonstrate that this classifier can get a high performance on infrared faceimage.The labeled graph is used to represent infrared face image, every node of the labeledgraph is labeled with a set of two-dimensional Gabor wavelets transform coefficients whichdescribe the local facial feature, and these nodes lie at the feature point positions of the faceimage which is useful for recognition; every edge of the labeled graph is labeled with metricinformation on the relative position of two adjacent nodes, grid structure that is composed byall edges describes the geometrical feature of the whole face. The elastic bunch graphmatching algorithm matches the infrared face image to a face bunch graph in order to obtainthe positions of the feature points, then the two-dimensional Gabor wavelets transformcoefficients are computed at the feature points and these coefficients are used for faceclassification and recognition.Three-lay BP Neural Network was designed based on the principle of Neural Network.The feature vector of infrared face image is extracted by the method of PCA+LDA, thefeature vector is used to training BP Neural Network, and a classifier with strong robust andfault tolerance can be generated. The infrared face recognition experiment was conductedwith the classifier and experimental results show that the method can get a high performance.