针对人脸图像数据量大和DBNs初始化权值具有指向性以及非凸目标函数易陷入局部最优的问题，提出将主分量分析（PCA）、遗传算法（GA）、深度信念网络（DBNs）相结合的新算法，并将其应用在人脸识别中。首先通过 PCA对人脸图像进行处理，以减小人脸图像的数据量，然后利用GA算法对DBNs进行逐层预训练以优化其网络权值，再利用BP算法对DBNs进行微调并构造分类器。以ORL数据库为实验数据通过与其他经典人脸识别算法的比较得出，该算法不仅可以减少人脸图像数据量，而且可以克服初始权值的指向性和局部最优问题，更重要的是可以提高识别精度和识别速度。%Aiming at the problem that the data volume of face image and DBNs initialization weight are pointed and the non convex objective function is easy to fall into local optimum , a new algorithm combining principal component analysis(PCA) ,genetic algorithm(GA) and deep belief network(DBNs) is proposed , w hich is applied to face recognition . Firstly , the face image is processed by PCA to reduce the amount of data . Then the GA is trained by DBNs algorithm to achieve the purpose of optimizing the weights . Then the BP algorithm is used to fine tune the DBNs and construct the classifier . According to the experimental results of ORL database , the proposed algorithm can not only reduce the amount of human face image data , but also overcome the initial weight of the directional and local optimization problems .More important ,it is to improve the recognition accuracy and convergence speed .
展开▼
