ENHANCING FACIAL EXPRESSION RECOGNITION THROUGH LOCAL DIRECTIONAL NUMBER PATTERNS: A NOVEL FEATURE DESCRIPTOR APPROACH

In this work, we present the local directional number pattern (LDN), a unique local feature descriptor, for face analysis, i.e., expression and face recognition. LDN produces a more discriminative code than existing techniques by compactly encoding the directional information of the face's textures (i.e., the structure of the texture). We calculate the structure of each micro-pattern using a compass mask that extracts directional information. Then, we use the well-known direction indices and sign to encode this information, allowing us to distinguish between related structural patterns with differing intensity transitions. We separate face into a number of areas and determine how the LDN characteristics are distributed throughout each. The result of combining these traits is a feature vector that serves as a face descriptor. This method concentrates on choosing localized characteristics from face expression photographs and classifying them using regression values, or partial F-test. In terms of robustness in appropriate feature selection and classification, the results demonstrate that conventional approaches are superior. By putting forth a reliable method called stepwise linear discriminant analysis, which concentrates on choosing the localized features from the activity frames and classifying them according to regression values, the most significant features were chosen. The objective of a feature extraction strategy is to derive from faces localized characteristics that prior feature extraction methods were unable to examine