Research And Application Of Portrait Search System Based On Deep Learning

With the rapid development of deep learning,the use of deep learning for portrait search has become a hot research direction.Face recognition and image search are two main research directions for searching information through face pictures.In various fields such as security,and monitoring,face recognition technology for the protection of people’s data privacy provides technical support.In this paper,portrait search method based on deep learning is used to search basic information through face images.The main work is as follows:1)Generate face detection model based on Dlib training.Firstly,a batch of photos with faces and non-faces are screened,and Img Lab is used to annotate photos with faces.Then,Dlib and Open CV tools are used to train face detection to form two face detection models.By comparing the detection results of the two models in the picture,it is shown that the model generated by face detection training based on Dlib can effectively detect the face region including the forehead.2)Perform face key point location based on the Mobile-HRNet network.In order to obtain high-resolution face key point location and reduce the number of parameters and network structure,this paper uses Mobile-HRNet combined neural network for face key point location.The error rate and failure rate of a single network such as Mobile Net V2＿0.25,Mobile Net V2＿1.00,HRNet V2 and Mobile-HRNet network for face key point location on the 300 W dataset were compared,the experimental results show that the accuracy of Mobile-HRNet network is higher than that of a single network.3)Portrait search based on face key point distance ratio.In order to solve the problem that face retrieval technology is time-consuming,and inefficient,and the features converted into face images can’t be explained,this paper uses the feature that the same person has similar proportions of facial features to convert face pictures into twenty-one dimensional features for portrait search.For face image set clustering,the shortest average time for searching each image is 0.138 s,while the shortest average time for portrait search based on face key point distance ratio is 0.136 s for searching each image.Experiments show that the algorithm proposed in this paper can ensure high accuracy and improve the speed of face image search effectively.