| Face alignment is mainly used to locate the key points of the face,that is,in the face area,describe the eyes,nose and other local key positions and contours.Face detection is the basis of face alignment,its main role is to find the face position in the image.Face alignment can give a more detailed facial feature point distribution,so it is widely used in face recognition,face animation and attitude estimation and so on.In the practical application,not only requires the algorithm to have high accuracy,but also need the algorithm have high speed.In order to apply the face detection and alignment algorithm to life better,this paper will improve the processing speed and accuracy of the algorithm as the main research content.The main research work in this paper has the following two aspects:1.In the face detection,the traditional Viola-Jones face detection algorithm(abbreviation: VJ face detection algorithm)to the video of each frame are the overall traversal to detect the face,in order to reduce the time spent by invalid traversal,The continuity and similarity between the video frames are improved,and the face position prediction algorithm is proposed.According to the number of faces,the algorithm divides the video into fewer face video and multi-face video,and proposes different face position prediction methods.The small face video uses the regression algorithm to predict the face of the video frame to be detected And then detect the face in the estimated range,so that the area where the required traversal is detected is reduced from the entire video frame to the predicted position range,thereby reducing the time consumption;the multi-face part uses the SVM response graph to track the video through VJ The algorithm detects the face target and derives the target position.Although the time complexity of the tracking algorithm is relatively high,it is still much lower than the time it takes to traverse the video as a whole.Finally,the two kinds of position prediction methods are integrated by the adaptive threshold equation given in this paper,which can be applied to any real-time video for face detection.2.In the aspect of face alignment,this paper optimizes the CLM algorithm framework.In the CLM algorithm,SVM is used to construct the response graph and fit the data of the response graph,which requires multiple iterations.This improves the accuracy of the algorithm,but also increases the time complexity of the algorithm.In order to reduce the time-consuming,this paper optimizes the shape model in the CLM algorithm model,uses the FPS3000 algorithm with fast speed but slightly reduce accuracy to replace the original average shape to reduce the number of iterations of data fitting in the CLM algorithm.This paper improves the algorithm toensure the accuracy of the algorithm at the same time,but also effectively reduce the time-consuming of algorithm.Through the algorithm which is proposed in this paper,the operation speed of the face detection part is improved by about 50%.The improved face alignment algorithm can process the face in the video in real time accurately.In addition to the performance comparison of the improved algorithm experiment,and finally also use the algorithm proposed in this paper to complete a real-time video face animation application case. |
