Face Recognition Under Changing Light Conditions

Face recognition，as a hotspot in the research of the field of biometric identification,involves the content of pattern recognition, machine learning, computer vision, imageprocessing, and so on. As a kind of non-contact, non-disturb biometric identificationtechnology, face recognition has a broad application prospect in the entrance guardsystem, monitoring, identification, etc. for the advantages of direct, concealment, etc. Atpresent, a lot of good performance face recognition algorithms have been proposed andsome successful commercial system also has been applied in reality. However, the facerecognition rate will be decline greatly, when faces with the inevitable problems,suchas illumination changes, posture, facial expression, ages, and so on, especially theillumination variation. Therefore, how to improve recognition rate and robustness underthe condition of changing light is a topic worth studying.In view of illumination variation problem in face recognition, illuminationcompensation and image illumination invariant feature extraction were studied in thispaper.For images under changing light conditions, the illumination compensation is themost important step in image preprocessing. Illumination compensation algorithm basedon pulse coupled neural networks was drew out through introducing the principle andadvantages and disadvantages of Several kinds of classic illumination compensationmethod with better effect in the third chapter; Basing on automatic potter characteristicsof pulse coupled neural network, combining the statistical features of images, using adiffusivity function related to the gradient and the gradient direction and defined byWeickert J, anisotropic pulse coupled neural network model is put forward and used forillumination compensation.The whole face local binary pattern (LBP) and the local facial organs LBP featuresare combined to implement feature extraction. Considering the primary and secondarydivision of facial features and the integrity of local characteristics, the whole LBPfeature is extracted, the main characteristic regions, such as nose, eyes, mouth, arebisected and LBP features of these regions are extracted at the same time. Later, theLBP features are connected and from comprehensive LBP feature. PCA algorithm isused to reduce dimension in order to reduce the amount of calculation. At last, thesimilarity of dimension reduction LBP features is calculated to classify face by2 statistical method.By the end of the article, in order to verify the applicability and effectiveness of theproposed method, the illumination compensation experiment is carried out on part ofimages in Yale B and MCU-PIE face database and the compensation effect is evaluated;The face recognition experiment on the images after illumination compensation usingthe proposed feature extraction methods; A real-time face recognition system isdesigned basing on OpenCV, realizing the collection of real-time face data, artificialsample selection, sample training and face recognition.