Answers are grounded in this paper and cite their sources.
Objectives: The main objective is to explore and develop an Intensity-Adaptive Local Binary Pattern (IA-LBP) handcrafted texture descriptor along with Sobel and Histogram of Oriented Gradient (HOG) feature fusion. This approach aims to enhance recognition of seven facial emotion classes such as anger, disgust, fear, happiness, sadness, surprise and neutral across diverse FER benchmark datasets. Method: The proposed IA-LBP handcrafted texture descriptor is integrated with Sobel and HOG via feature fusion analogy. IA-LBP determines adaptive threshold by using neighbourhood mean intensity and Shannon entropy. Its 256-bin histogram is integrated with six statistical Sobel features and 900 HOG features resulting in 1,162-dimensional representation from facial image. The fused feature vector is standardized using Z-Score normalization. Subsequently, SMOTE strategy is employed to address class imbalance. Further emotion prediction on facial images is carried out using various classifiers namely, Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT) and K-Nearest Neighbour (KNN). Performance evaluation is done using stratified five-fold cross-validation on five FER benchmark datasets. The efficacy of the model is evaluated using standard classification metrics like accuracy, precision, recall and F1-score. Findings: In the proposed IA-LBP with Sobel and HOG framework, SVM recorded the highest accuracy on CK+ (99.77%), JAFFE (98.48%), FER-2013 (82.19%) and RAF-DB (91.46%), while KNN recorded the highest accuracy on KDEF (92.00%). The effectiveness of the proposed framework is demonstrated across controlled and unconstrained facial expression datasets. The results were reported with accuracy gains of 0.17, 6.68, 1.80, 5.99 and 41.96 percentage points over previous LBP-based methods on the CK+, JAFFE, KDEF, FER-2013 and RAF-DB FER bench mark datasets respectively. However, these gains should be interpreted cautiously, taking into account differences in experimental protocols used in the compared studies. Novelty: The study evaluates a hybrid model by fusing IA-LBP with Sobel and HOG features. This technique substantially improves facial emotion recognition by employing the adaptive texture encoding in combination with complementary gradient features.
No takes yet. Share an insight, caveat, or question.
Poornima et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: