引入图像去噪处理机制的DNN模型人脸识别算法

Face Recognition Algorithm Based on Deep Neural Networks Model Introducing Image Denoising Processing Mechanism

为提高人脸图像特征描述的完整性、准确性和精细性,本研究提出一种基于去噪处理和DNN模型应用的人脸识别算法。在DNN模型的神经元激活函数和损失函数中引入了去除人脸样本自身存在和处理过程产生噪声的处理机制,采用增加模型神经元数量控制启动层数、将全连接层与分类匹配识别层合并优化模型结构,通过定义辅助参数,自主实现模型学习训练的权值和偏值参数调整,以确保本研究算法的运行效率,提高人脸识别的可靠性和稳定性。通过实验自证和与其他算法的比证,表明本研究算法对人脸特征提取的准确率达到92%以上,人脸匹配识别的准确率达到95%以上,系统运行占用CPU的处理时间在13s以下,具有一定的优势。

In order to improve the completeness, accuracy and refinement of facial image feature description, in this study, a face recognition algorithm was proposed based on denoising processing and the application of DNN model.In the neuron activation function and loss function of the DNN model, a processing mechanism was introduced to remove the existence of face samples and the noise generated by the processing process. The number of neurons in the model was increased to control the number of startup layers, and the fully connected layer was combined with the classification, matching and recognition layer. The model structure was optimized, and the weights and bias parameters of the model learning and training could be adjusted independently by defining auxiliary parameters,so as to ensure the operation efficiency of the algorithm in this study and improve the reliability and stability of face recognition. Through experimental self-evidence and comparison with other algorithms, it showed that the accuracy rate of the algorithm in this study for face feature extraction is over 92%, the accuracy rate of face matching recognition is over 95%, and the processing time of the system running on the CPU is 13s below, there are certain advantages.