Facial Emotion Recognition Based on Local Feature

Facial expressions are categorized into seven types: happiness, sadness, fear, disgust, anger, surprise and neutral. Due to factors like uneven lighting and minute expression variations, current facial emotion recognition algorithms often struggle with accuracy. To address this, this paper presents a novel approach to recognizing facial emotions. Firstly, applying Gamma correction to homogenize lighting in facial images, enhancing lighting conditions for precise recognition. Secondly, using anthropometric methods to precisely locate and segment key facial areas like eyebrows, eyes and mouth, and extract the primary emotion-reflecting areas. After that, a feature plane is created based on these facial emotion components, so as to eliminate redundant data from non-emotional facial areas to boost recognition efficiency and accuracy. Lastly, deep facial features are extracted using a convolutional neural network to complete facial emotion recognition. Experiments demonstrate the efficacy of the constructed feature plane and a notable improvement in recognition accuracy over geometric feature-based methods, achieving a 90.12% accuracy rate on the JAFFA public expression datasets.