Unmasking Emotions: Deep Neural Networks for Image-Based Emotion Recognition

Recognizing human emotions is a multifaced field that uses the technology, especially deep learning and Convolutional neural networks, to recognize and understand emotions such as happiness, sadness, anger, disgust and fear. Convolutional Neural Networks are designed for visual data explorations and they are effectively capture facial expressions for emotion recognition. The process involves the collection of labeled data, and also preprocesses it, creating an appropriate deep neural network, training the model with constant updates, and assessing its performance using metrics on a testing dataset. Proposed work tackles emotion detection in irregularly distributed datasets of preprocessed 48x48 grayscale emotion images. Significant performance gains are achieved through targeted data augmentation for marginalized classes, giving valuable insights for robust emotion recognition models. Optimizing hyperparameters enhances the model, enabling its deployment in real-world scenarios by ensuring integration, data handling, and real-time functionality. The proposed model primarily focuses on offline image and video capturing for emotion detection, and highlighting on video-emotion recognition.