基于卡方距离度量学习的面部表情识别算法

Facial expression recognition algorithm based on chi-squared distance metric learning

针对野外复杂环境下面部表情特征不一致导致识别率低的问题,提出一种基于卡方距离度量学习的凸优化算法用于面部表情识别。将卡方距离引入KNN分类技术中用于度量学习优化的损失函数,采用随机梯度下降法求解修正的凸优化损失函数,为避免过度拟合训练数据,算法将Dropout技术用于度量学习,使用特征权重系数,调整不同特征对表情识别的贡献度。实验结果表明,相比其它算法,所提算法在面部表情识别中更具优势,提高了面部表情识别准确度。

Aiming at the problem of low recognition rate caused by inconsistent expression features in the complex environment of the wild,a convex optimization algorithm based on chi-squared distance metric learning was proposed for facial expression recognition.Chi square distance was introduced into KNN classification technology to measure the loss function of learning optimization,and random gradient descent method was used to solve the modified convex optimization loss function.The feature weight coefficient was used to adjust the contribution of different features to expression recognition.Experimental results show that compared with other algorithms,the proposed algorithm is more advantageous in facial expression recognition,improves the accuracy of facial expression recognition.