Facial Emotion Recognition Using Local Directional Ternary Patterns and SVM

Facial Expression Recognition (FER) is a crucial component of human-computer interaction, affective computing, and behavioural analysis systems. Although deep learning-based methods have shown impressive results, they are often data-intensive and computationally expensive. In contrast, handcrafted texture descriptors remain useful for FER tasks with limited data. In this paper, we present an improved Local Directional Ternary Pattern (LDTP)-based facial expression recognition system that aims to enhance directional texture pattern encoding and spatial feature aggregation for better discriminative power. The proposed method focuses on the major directional patterns and combines spatially localized feature description over facial regions, allowing for better expression-related subtle variations. The LDTP features are then modelled using a Support Vector Machine classifier with a radial basis function kernel. Experiments on the CK+ database with an image-level stratified train-test split show that the proposed method can achieve an accuracy of 89.34%, which is better than the baseline LBP- and traditional LDTP-based methods under the same experimental setup. These findings suggest that the proposed LDTP-based approach is an effective and efficient solution for facial expression recognition.