Enhancing Facial Expression Recognition Through an Innovative Feature Selection Technique Based on Fisher Criterion and Genetic Algorithm

Facial Expression Recognition (FER) is a pivotal and captivating field in computer vision with broad applications that have made it a popular research topic since the 1990s, playing a significant role in image processing. FER typically involves two key stages: feature extraction and selection, followed by classification. This study introduces a novel feature selection approach, named FCGA (Fisher Criterion and Genetic Algorithm), to address challenges in FER. In the FCGA method, the Fisher criterion is used to evaluate feature subsets, and a genetic algorithm is devised to identify an optimal feature subset from the pool of candidate. The FCGA method is applied to select features for FER from various types, such as Local Phase Quantization (LPQ) and Pyramid of Histogram of Oriented Gradients (PHOG). These selected features then collaborate with Support Vector Machine classifiers to categorize facial expressions. Experimental results utilizing the JAFFE and FER2013 databases reveal that the FCGA method boosts the recognition rate by a noteworthy margin of up to 1.3% and 3.2% compared to other feature selection methods.