Research On Face Mosaic Processing Technology

With the development of technology,face recognition is increasingly used in social life,so everyone's facial information is more and more important.However,some of the current press photos fail to protect the face information involved,so you need to study a method that can perform face mosaic processing on the captured pictures.Face mosaic processing is to select the position of the face in the photo and then mosaic it to protect the face information,so accurate detection of the face is the key step.Face detection belongs to the category of target detection.Among the two methods of target detection,the two-stage face detection method has always been very hot since the beginning of the study because of its high accuracy.This article uses a two-stage method to detect faces in photos.The main research contents of this article are as follows:(1)Compared with the two mainstream candidate box algorithms,a high recall selection search algorithm is used to extract candidate boxes from the original photos.Then the size of the input picture is improved according to the characteristics of the algorithm,thereby speeding up the extraction speed of the candidate frame and reducing the number of candidate frames.Through a large number of experiments,it is found that the number of candidate frames cropped to include faces after the algorithm is improved changes little,and the face appearing in the cropped image accounts for a relatively large amount,which meets the requirements of the subsequent neural network two classification.(2)A new network structure has been improved based on the Le Net5 network.Increased the number of convolutional layers on the original structure,reduced the number of fully connected layers,and added Dropout to prevent overfitting;Specifically,the 5×5 convolution kernel is replaced with a 3×3 small convolution kernel,and the number of convolution kernels in each layer is increased to change the activation function..Finally,the same data set is used to train the original Le Net5 and VGG11 networks.The comparison results show that the improved network has good performance,the accuracy of face detection is greatly improved compared to the basic network,and the amount of parameters is reduced by three-thirds compared to VGG11.Second,it can detect the faces in the photos verywell.At the same time,a new coding method is designed,the speed is improved compared with the average coding.In this paper,the face mosaic method based on the selection search kernel to improve the convolutional neural network has a better coding rate to meet the needs of coding,and it can well perform face detection and mosaic processing on a single picture.At the same time,the whole algorithm has good stability and low complexity,and can be well used in the processing of face mosaic in photos.