Basic Pose Classification of Pendet Bali Dance Images Using Multilayer Perceptron (MLP) Based on GLCM Feature Extraction and Otsu Thresholding

Traditional dance classification presents a significant challenge due to the complexity of body basic poses and visual similarities across gestures. This study aims to develop an intelligent system for classifying Pendet Balinese dance basic poses using artificial intelligence and image processing techniques. The research applies a quantitative experimental approach combining Otsu Thresholding for segmentation, Gray Level Co-occurrence Matrix (GLCM) for feature extraction, and Multilayer Perceptron (MLP) for classification. A total of 690 labeled images from 15 distinct Pendet basic poses were collected from professional dancers and preprocessed into binary form using Otsu’s method to isolate the dancer from the background. Subsequently, GLCM features, energy, contrast, correlation, and homogeneity, were extracted across four directions. These features served as input for the MLP classifier, trained using a 10-fold cross-validation technique. The model achieved an overall classification accuracy of 82.75%, with high precision and recall for several basic pose types such as Ngelog, Ngeseh, and Nyeregseg. However, some basic poses with overlapping poses presented classification difficulties. The confusion matrix analysis indicated the model's capacity to differentiate most basic pose types effectively. This research demonstrates the feasibility of using MLP combined with texture-based feature extraction for cultural motion classification. The findings contribute to the digital preservation of traditional Indonesian dance and offer a foundation for developing educational and archival tools. Future improvements may involve the use of deep learning techniques and temporal video data for enhanced performance.