Design And Implementation Of Face Analysis System Based On The GPU Acceleration

In recent years, face image analysis has been a hot topic in the field of computer vision and pattern recognition. In human-computer interaction, biometrics, visual surveillance, and multimedia fields, there is a wide range of applications. Face Detection and face alignment is the most basic tasks, but also one of the key tasks of facial images analysis. In this paper, the GPU parallel computing features are incorporated on face detection and localization of facial key points. Based on the analysis and summary of the research progress at home and abroad in recent years, we study the AdaBoost face detection algorithm based on Haar features and localization of key points using the active shape model(ASM). We excavate parallel point on the conventional algorithms and use GPU to accelerate the algorithms. For face detection, the calculation of inter-grogram, scanning window of detection and correction of window to merger were used parallel points. The use of GPU accelerated parallel experimental results on FDDB database show that the proposed method in terms of speed upgrade has obvious advantages. For face alignment, the paper will partially improve the classical ASM model algorithm by using more expressive SIFT feature descriptions, and seeking local image descriptors in parallel way. Extensive experimental results show that this method is not only fast, but also high accuracy. Experiments on LFPW and Helen database show that the proposed method has better results. On the basis of face detection and face alignment, inspired by psychological research, this paper proposes the joint coding method to describe the characteristics of the facial expression. Boosting is used based on a joint Haar features to generate classifier. Experimental results on CMU and universal facial expression database JAFFE show that the proposed method can obtain a better recognition performance. For face attribute analysis, we extract Gabor features of the face, SVM classifiers are trained to analysis race, gender, wear glasses or not, has beard or not, etc. In the actual application of this attribute analysis algorithm, we can get the good results. Face detection, face alignment, facial expression recognition, facial attribute analysis methods proposed with GPU acceleration on this paper show the good performance in our facial analysis system. Facial analysis system constructed in this paper has a high practical value, both in the PC platform or mobile platform, the results are well worth promoting.