Facial Expression Recognition Using Inception V3-Based FeatureExtraction and SVM Classification on the Affectnet Dataset

Introduction: In the field of human–computer interaction, in the context of affective computing, as well as behavioural analysis, Facial Expression Recognition (FER) is usually used as an important technique. Convolutional Neural Networks (CNNs) are widely applied, and they have features such as ResNet, VGG, and Inception. materials and methods: This study suggests a hybrid strategy that combines the inception-v3 model (deep feature extraction) with Support Vector Machine (traditional machine learning) to classify facial emotions effectively. To extract high-level feature representations from facial images, in this paper, we include a hybrid of Inception-v3 and an SVM classifier with the AffectNet dataset for facial expression recognition. We consider linear SVM as well as RBF SVM and make a comparison between them. Materials and Methods: In this study, a hybrid method integrates Inception V3 with a Support Vector Machine (SVM), a traditional machine learning approach. In this paper, these very clear details of the face are obtained from the images with the application of the Inception V3 model. results: This hybrid achieves 78 Results: The results of this simple experiment show an accuracy of 78% for two different emotions, 60% for three different emotions, and 37% for seven different emotions, when the RBF SVM classifier is used. Similarly, when the linear SVM classifier is applied, the model achieves 73% accuracy for two different emotions, 55% for three different emotions, and 32% for seven different emotions. Discussion: From the results, it can be seen clearly that the SVM classifier works very well when it is used alongside the Inception V3. In this study, we understand how effective the feature extraction–based pipeline method is, which is used instead of the bigger full fine-tuning model. conclusion: This framework presents a scalable and effective solution for real-time emotion recognition systems and intelligent applications. Inception V3 removes its final classification layers and extracts high-level facial features from preprocessed images, and an SVM classifier (both linear and RBF kernels) is used for emotion classification. The AffectNet dataset is used to train and evaluate the model. Conclusion: In this study, we make a comparison between linear SVM and RBF SVM classifiers using deep features from the V3. The results show that the model achieves higher accuracy and macro-F1 scores when compared to the linear SVM.