小规模数据多角度人脸表情人工智能识别仿真

Small-Scale Data Multi-Angle Facial Expression Artificial Intelligence Recognition Simulation

已有人脸表情智能识别方法存在图像遗漏、识别速度慢、以及表情易混淆的问题,导致识别正确率低,且不适用于多人脸识别。为解决上述问题,提出小规模数据多角度人脸表情人工智能识别方法。结合去噪自编码器、稀疏自编码器及普通自编码器组成具有5层网络结构的堆栈式混合自编码器。在网络结构中训练数据样本,并对样本完成微调、权重初始化和更新。再结合粒子群优化分类器识别表情,将粒子最优值带入人脸表情识别目标函数,获得判定表情的决策矩阵,完成人脸表情人工智能识别。仿真结果表明,所提方法识别多人表情识别时无遗漏,且识别速度快、精度高、人脸表情不易混淆。

In this paper,an artificial intelligence recognition method of multi-angle facial expressions based on small-scale data was proposed.Denoising Auto-Encoder(DAE),Sparse Auto-Encoder(SAE)and Ordinary Auto-Encoder(OAE)were used to construct a Stacked Hybrid Auto-Encoder(SHAE)with a five-layer network structure.Data samples were trained in this network structure.Moreover,these samples were fine-tuned,weight-initialized and updated.A particle swarm optimization classifier was used to recognize expressions,the optimal particle value was introduced into the objective function of facial expression recognition,so that the decision matrix to determine expression was obtained.After that,the artificial intelligence recognition of facial expressions was completed.Simulation results prove that the proposed method has no omission when recognizing multiple persons'expressions,with high recognition speed and accuracy.Moreover,the facial expressions are not easy to be confused.