EXPERIMENTAL ANALYSIS OF FACE RECOGNITION METHODS AT VARYING SENTIMENT CASES

Face recognition has been extensively studied using curated datasets, where multiple images of each individual are typically captured under consistent emotional expressions.As a result, most experiments rely on uniform data, overlooking the challenges posed by varying sentiments.However, in real-world applications, individuals often display a wide range of emotions, making it essential to evaluate face recognition methods under such diverse conditions.Addressing this problem is crucial for developing robust biometric systems capable of maintaining accuracy despite emotional variations.This study investigates the performance of face recognition techniques when individuals exhibit significantly different emotional expressions.Specifically, we evaluate three feature extraction methods-Histogram of Oriented Gradients (HOG), Local Binary Patterns (LBP), and raw pixel values-using a k-Nearest Neighbors (k-NN) classifier.By leveraging a dataset of 20 individuals expressing eight distinct sentiments, we aim to determine the robustness and accuracy of each method under emotionally diverse conditions.Experimental results demonstrate that the Local Binary Patterns (LBP) method consistently outperforms the HOG method in scenarios involving varying sentiments.This superior performance can be attributed to LBP's ability to capture fine-grained texture patterns that remain consistent despite changes in facial expressions.In contrast, HOG, which primarily encodes shape and edge information, shows reduced effectiveness when facial contours shift due to emotional variations.The findings of this study have significant implications for real-world face recognition systems, particularly those deployed in dynamic environments where emotional diversity is common.By highlighting the advantages of LBP over HOG under these conditions, this research contributes to the development of more resilient biometric authentication solutions.