Research On Face Recognition Algorithm Under Complex Conditions

Currently,face recognition technology is widely used in various aspects such as regional security and identity verification.Detecting and boxing out the complete facial area is a prerequisite for face recognition.Under the controllable condition,each person can obtain a plurality of facial images in different scenes,so that during training,we can fully learn the facial recognition feature information of a certain person.However,under some uncontrollable complex conditions,such as inconvenient collection or limited storage,each person has only one facial image,and there may be changes in illumination,expression,occlusion,and posture between the test sample and the training sample.Due to the insufficient number of training samples,these changes can not be fully learned during the training process,and the recognition performance of the multi-sample face recognition algorithm will be seriously degraded.SRC and CRC are algorithms that are widely used in the field of image classification.This paper mainly uses SRC and CRC to study the single-sample face recognition algorithm in the presence of face changes such as illumination,occlusion,expression and posture.The specific work is as follows:Firstly,the multi-task cascade neural network is used to detect the face and face area.The algorithm is composed of three neural network layers cascaded,which can complete the face/non-face detection and face boundary from coarse to fine.The regression of the box determines the three tasks of positioning the key feature points of the face.Secondly,the SRC and CRC classification algorithms are introduced.Through multiple sets of experiments,it is found that these two algorithms have high classification accuracy under sufficient training samples,and CRC has better classification speed than SRC.Trying to construct more training samples is a great idea of applying CRC to single sample face recognition successfully.Thirdly,using the similarity between different face images,the changes of illumination,occlusion and expression of the face are constructed from the additional generic dataset to make up for the lack of information of the single training sample.Face local information also plays an important role in face recognition.In this paper,a single-sample face recognition algorithm combining global and local common representation is proposed.This algorithm is a dual classification mechanism.Firstly,the global error function and the adaptive change dictionary are constructed by global cooperative representation;then all the samples are divided into blocks,and the block generic cooperative representation is performed,and the block classification is corrected and adjusted by the global error function;finally,according to all the block classification results to get the test category.The simulation experiment on the face database verifies that the proposed algorithm has excellent performance.Finally,the generic learning method cannot learn the nonlinear facial variations such as expressions and postures.We introduce the key point positioning block and Gaussian mixture model in the generic cooperative representation.The image is segmented based on the position coordinates of the five key points of the face to solve the problem of block dislocation caused by the face deflection.The Gaussian mixture model(GMM)is used to iteratively cluster the mixed data to eliminate the face.The effect of linear changes on recognition rate.The experimental results verify that the proposed algorithm has high recognition performance.