An Out-of-Distribution Attack Resistance Approach to Emotion Categorization

Deep neural networks are a powerful model for feature extraction. They produce features that enable state-of-the-art performance on many tasks, including emotion categorization. However, their homogeneous representation of knowledge has made them prone to attacks, i.e., small modification in train or test data mislead the models. Emotion categorization can usually be performed either in-distribution (train and with same dataset) out-of-distribution one more dataset(s) different dataset). Our already developed landmark-based technique, which is robust improvement against attacks categorization, could translate classification problems. This important as databases might have variations such color level expressiveness emotion. We compared method four deep models (EfficientNetB0, InceptionV3, ResNet50, VGG19), well tools (i.e., Python Facial Expression Analysis Toolbox Microsoft Azure face application programming interface) by performing cross-database experiment across six commonly used databases, extended Cohn–Kanade, Japanese female facial expression, Karolinska directed emotional faces, National Institute Mental Health Child Emotional Faces Picture Set, real-world affective psychological image collection at Stirling databases. The achieved significantly higher accuracy, achieving an average 47.44% most ( $< $ 36%) 37%) considerably less execution time. highlights much harder task due detecting underlying cues than where superficial patterns detected notation="LaTeX">$>$ 97% accuracy. Impact Statement—Recognising emotions from people's faces applications computer-based perception it often vital interpersonal communication. recognition tasks nowadays addressed using learning colour distribution so classify images rather contrast hypothesised heterogeneous landmark-based. investigated through problems, samples drawn dataset training images. achieves (on average) ResNet50 other Py-Feat Face API. conclude this improved generalization relevant future developments tools.