基于深度学习混合模型的人脸检测算法

Face Detection Algorithm Based on Mixed Model of Deep Learning

基于“单一深度模型”的人脸检测算法在人脸图像存在部分遮挡情况时可能会导致学习效率低、错误检测率高,因此笔者提出了一种基于深度学习的混合模型算法解决人脸检测中存在的问题,称为CPDBN模型(卷积池化深度置信网络)。首先,将卷积神经网络的池化层和卷积层添加到受限玻尔兹曼机的隐含层中,作为基本单元深度学习的主要内容。其次,结合深度模型的深度结构应用特点构建多层基本单元结构,分析描述人脸特征的不同。最后,当分析过程受到阻碍,如人脸被遮挡等情况时,则以完整人脸特征作为检测参考进行特征分析。根据本实验结果,该算法加快了收敛速度,提高了局部遮挡时人脸检测的精度,提高了多姿态的鲁棒性。

Face detection algorithm based on"single depth model"may lead to low learning efficiency and high error detection rate when face image has partial occlusion.Therefore,the author proposes a hybrid model based on deep learning to solve face detection.The problem is called CPDBN model(convolution pooled deep belief network).First,the pooling layer and convolution layer of the convolutional neural network are added to the hidden layer of the restricted Boltzmann machine as the main content of the basic unit deep learning.Secondly,combined with the deep structure application characteristics of the depth model,the multi-layer basic unit structure is constructed,and the differences in facial features are analyzed and described.Finally,when the analysis process is hindered,such as when the face is occluded,the feature analysis is performed using the complete face feature as a detection reference.According to the experimental results,the algorithm speeds up the convergence,improves the accuracy of face detection during local occlusion,and improves the robustness of multi-pose.