Device-Based Majapahit Inscription Classification with Multi-Filter Enhancement

Abstract

The preservation of cultural heritage through digitalization has become increasingly essential in modern archaeological and information technology research. This study focuses on classifying Majapahit inscription images based on the recording device using machine learning approaches enhanced by multiple image filtering techniques. A dataset comprising seven inscriptions photographed with seven different devices was used to evaluate the performance of three classification models: Logistic Regression, Support Vector Machine (SVM), and K-Nearest Neighbors (KNN). Four preprocessing filters Grayscale, Sobel, Histogram Equalization, and Canny Edge Detection were applied to assess their effects on model accuracy. The results revealed that the SVM consistently achieved the highest accuracy and robustness, particularly under Sobel and Histogram Equalization filters, confirming its superior ability to capture discriminative texture and edge-based features. In contrast, KNN showed unstable results due to its sensitivity to noise and intensity variations, while Logistic Regression performed moderately well in linearly separable data conditions. Paired t-test analysis further validated that SVM’s performance advantage was statistically significant. These findings highlight that edge-preserving preprocessing techniques can substantially enhance the accuracy of device-based image classification and provide a computational framework that supports digital preservation efforts in cultural heritage research.