一种特征加权融合人脸识别方法

针对传统人脸识别算法在非限制条件下识别准确率不高的问题,提出了一种特征加权融合人脸识别方法(DLWF⁺)。根据人脸面部左眼、右眼、鼻子、嘴、下巴等5个器官位置,将人脸图像划分成5个局部采样区域;将得到的5个局部采样区域和整幅人脸图像分别输入到对应的神经网络中进行网络权值调整,完成子网络的构建;利用softmax回归求出6个相似度向量并组成相似度矩阵与权向量相乘得出最终的识别结果。经ORL和WFL人脸库上进行实验验证,识别准确率分别达到97%和91.63%。实验结果表明:该算法能够有效提高人脸识别能力,与传统识别算法相比在限制条件和非限制条件下都具有较高的识别准确率。

Face recognition by weighted fusion of facial features

The accuracy of face recognition is low under unconstrained conditions. To solve this problem, we propose a new method based on deep learning and the weighted fusion of facial features. First, we divide facial feature points into five regions using an active shape model and then sample different facial components corresponding to those facial feature points. A corresponding deep belief network(DBN) was then trained based on these regional samples to obtain optimal network parameters. The five regional sampling regions and entire facial image obtained were then inputted into a corresponding neural network to adjust the network weight and complete the construction of sub-networks. Finally, using softmax regression, we obtained six similarity vectors of different components. These six similarity vectors comprise a similarity matrix, which is then multiplied by the weight vector to derive the final recognition result. Recognition accuracy was 97% and 91.63% on the ORL and WFL face databases, respectively. Compared with traditional recognition algorithms such as SVM, DBN, PCA, and FIP+LDA, recognition rates for both databases were improved in both constrained and unconstrained conditions. On the basis of these experimental results, we conclude that the proposed algorithm demonstrates high efficiency in face recognition.