Detection And Recognition Of Sensitive Face Images Based On Deep Learning In Video Stream

With the rapid development of the Internet and the continuous optimization of face recognition algorithms,face recognition technology is becoming more and more popular in the field of biometric recognition.In video detection,the difficulty of face detection and recognition is greatly improved due to the constantly changing face posture,and the influence of image background interference and uneven illumination.The effect of face recognition based on the traditional image recognition method is often not ideal.The rise of artificial intelligence and the rise of deep learning bring new opportunities for image recognition,and also bring face recognition to a new wave.In this paper,based on the deep learning method and the image processing technology,the face detection and recognition for the specific target characters in the video is realized.First,the video key frame is extracted from the video sequence,and then the face detection and recognition are carried out.In the phase of extracting video key frames,the detection and recognition of specific facial image based on the color characteristics of the image to represent,the key frame is extracted by comparing similarity of image gray histogram;the similarity of the video image frames are compared by the dichotomy,which improves the efficiency of the algorithm and time complexity.In the face detection,first of all the extracted key frame images are preprocessed by noise reduction,filtering and image enhancement.Because the complex background and uneven illumination of the traditional Adaboost face detection classifier based on Haar-like gray feature training affect the detection effect.In this paper,using the YCbCr color space to construct color Gauss model to filter the background and obtain the candidate skin area.Meanwhile,a skin brightness model is established to reduce the influence of uneven illumination.Then trained Adaboost face detection classifier is applied to face detection in candidate skin regions.Through a large number of control experiments,it is found that the face detection algorithm used in this paper has a higher positive detection rate.In the face recognition,we design a convolutional neural network structure model which is suitable for the specific face recognition based on the TensorFlow deep learning framework.The convolution neural network structure designed in this paper has 7 layers,which is different from traditional convolutional neural network.Convolution layer and pool layer are not alternating,but continuous convolution operations,which can extract higher dimensional features.And the local response normalization layer is added after the convolution pool operation,which can effectively improve the accuracy in the training process of deep learning.The model is tested on the data set collected in this paper,with a recognition rate of 89%,which effectively achieves the goal of detecting and recognizing sensitive face images in videos.