Facial expression classification based on LDA-GS-SVM
Fanchen Zheng · 2023
With the development of artificial intelligence (AI), facial expression recognition has become a focus of current research owing to its wide application potential. In this study, based on the karolinska directed emotional faces KDEF face public dataset, we adopted a parameter optimization method to optimize the kernel function parameters of the support vector machine which was used for classification. This study also conducted a comprehensive comparative analysis of the effects of expression recognition before and after parameter optimization. First, traditional manual and deep learning methods were used to extract face features, form two feature sets, and combine them to obtain a hybrid feature set. Second, linear discriminant analysis was used to reduce dimensionality, and a correlation matrix was used to select the class vectors to remove redundant features. Subsequently, grid search and k-fold cross-validation methods were used to optimize the parameters. The kernel function with default parameters and the parameter-optimized kernel function were used to classify the feature sets, and the effects of parameter optimization on facial expression classification were compared and analyzed. Finally, the classification effect of support vector machines with different parameters was evaluated. The results show that the overall classification effect of support vector machine is good and that parameter optimization can improve the accuracy of expression classification to a certain extent. The classification accuracy of the test set for the parameter-optimized support vector machine can be as high as 98.8%, indicating good applicability to the dataset.