Enhancing Face Emotion Recognition through LBP, SIFT boosted CNN Classification

Emotions are crucial because they allow wordless communication, guide decisions based on values, and support social connections. Face Emotion recognition (FER) technology has gained significant importance in various sectors, like human-computer interaction, healthcare, entertainment, etc. This technology enables machines to interpret human emotions from facial expressions, transforming interactions and experiences. This paper delves into FER methodologies, particularly focusing on refining accuracy through feature extraction by integrating Local Binary Pattern (LBP), Scale-Invariant Feature Transform (SIFT) with Convolutional Neural Networks (CNNs) and with only CNN. CKplus and FER2013 datasets are used for evaluation. The primary finding is that the LBP-CNN model was more accurate than SIFT-CNN (90.86%) and CNN without feature extraction (98.98%) on the CKplus dataset, with a 99.49% accuracy. On the FER2013 dataset, CNN without FE was 52.53% more accurate than both LBP-CNN and SIFT-CNN. This study demonstrates the potential of feature extraction methods with and without deep learning model by considering the impact of dataset complexity.