The Research Of Algorithm Of Face Detection And Recognition Based On Video

Face detection and recognition algorithm based on video research is aimed at today the era of big data,in the face of the video stream this large amount of data and the unconstrained condition of face recognition under complex scene is studied,which based on the three key links of face recognition:face detection,the key feature points localization and the face recognition feature extraction,the main research content is as follows:1.Research on face detection method based on pixel different feature.For all kinds of complicated unconstrained video scene,the different feature between the pixels is defined as a new face detection target feature,and take advantage of the pixel level features relevance information to simplify the model.At the same time in order to further improve the accuracy of classification model,on the basis of the classification of traditional single threshold segmentation with double threshold decision binary tree,using more advanced features categories of segmentation,the classifier so that you can better deal with complex light,variable natural scene,angle and occlusion,etc.2.The key feature points detection based on Ensemble of Regression Trees.Make full use of the iterative regression tree update properties,at the same time,using Gradient Boosting the method of regression tree pruning optimization,constantly according to the shape of the face feature of the current input to output estimate face shape analysis,and then by calculating each iteration of the updated value to update the next iteration of face shape input value,until the end of the iteration key feature points needed for positioning in the end.3.Research on face recognition method based on the deep learning.Through the study of deep learning method,in the related on the basis of the classical convolution neural network framework,by choosing different positions,different scales,and different color channel,even flip horizontal face patch optimization of network frame structure,layout and each layer of network parameters,the features of the extraction of face information have more comprehensive expression ability,at the same time the introduction of face recognition and face verification double supervision signal to ensure the accuracy of face recognition,and finally uses greedy algorithm for screening the most expressive part face patch to connect as the extraction of the facial features,makes face recognition in guarantee a high accuracy at the same time also has a good real-time performance.