A Lightweight Hybrid GLCM–MobileNetV2 Model for Batik Motif Recognition in Digital Cultural Learning Environments

Abstrak

Background: Automatic identification of Surakarta Parang batik motifs presents significant challenges due to the high visual similarity among sub-motifs, where conventional Convolutional Neural Network (CNN) architectures often fail to capture fine-grained texture characteristics.Purpose of Study: A lightweight hybrid model is proposed to integrate 24 GLCM-derived texture features with MobileNetV2 spatial descriptors through a feature fusion strategy to improve motif classification accuracy.Methodology: The proposed methodology employs a hybrid feature extraction strategy, where 24 texture descriptors consisting of six statistical parameters (Contrast, Correlation, Homogeneity, Dissimilarity, ASM, and Energy) calculated across four orientations (0°, 45°, 90°, and 135°) with 1,280 deep spatial features obtained from the MobileNetV2 backbone.Main Findings: Experimental results demonstrate that the proposed hybrid model achieves an accuracy of 99%, representing a substantial performance gain over the baseline MobileNetV2 model (66.67%) and the GLCM-SVM approach (85%). These results indicate that the integration of statistical texture descriptors and deep spatial features notably enhances the recognition of complex batik patterns. Furthermore, the findings suggest that this feature fusion approach is highly effective in resolving the intricate geometric similarities of Parang sub-motifs, providing a more reliable and efficient alternative to standard deep learning models for fine-grained classification tasks.Novelty/Originality of This Study: The novelty of this study lies in the implementation of a feature fusion strategy that compensates for the limitations of lightweight CNNs in texture recognition by incorporating classical statistical descriptors, specifically tailored for the intricate patterns of Parang batik.