Face Detection And Recognition Algorithm Research In Low Light

With the advent of the age of computer intelligence,face detection and recognition technology has attracted more and more attention as a popular direction in the field of computer vision.Face detection and recognition technology has broad application prospects and huge potential commercial value.However,the development of this technology is restricted by many factors,and low light is one of them.How to improve the face detection and recognition rate in low light environment is one of the important research directions at this stage.Therefore,this article mainly focuses on the problem of low illumination in face detection and recognition.The main research contents are as follows:(1)Research on low-light image enhancement.The classic histogram equalization,gamma transform and other algorithms are studied,and Analyze strengths and weaknesses.On the basis of the above,a low-light enhancement idea based on the combination of global brightness adjustment and global information is proposed.First,by establishing a global mapping function,the brightness of the dark areas of the image is improved as a whole,and the dynamic range of the image is compressed.At the same time,a proportional limit is placed on the magnitude of the brightness enhancement to prevent it from being excessively enhanced.On the basis of brightness enhancement,the relationship between the central pixel and the domain pixel is used to construct a contrast enhancement transfer function to enhance the local information contrast.Experiments show that this algorithm can achieve the purpose of enhancing the fine details of the image while increasing the brightness of the dark part of the image.(2)Aiming at the research of face detection and recognition algorithms in unconstrained environment,this paper proposes a face detection and recognition based on Tensorflow framework,which uses multi-task convolutional neural network and FaceNet model fusion.The advantage of the multi-task convolutional neural network is that it can perform face detection and face key point positioning at the same time,and filter out the vast majority of non-target areas in the image at the beginning of the detection,which greatly reduces the calculation and calculation of the network time.The FaceNet model adopts an end-to-end learning method,learningan encoding method from images to Euclidean space,and expressing the similarity of human faces through spatial distance.Finally,low-light enhancement is added to the combined model as a comparison.In order to verify the performance of the model,relevant experiments were carried out.Experimental results show that the combined model algorithm has certain robustness in unconstrained scenes.After adding the image enhancement model to enhance low-light images,it can effectively improve the face detection rate,which proves the effectiveness and reliability of the algorithm.