Document Deterioration Classification Using Gabor Filters and Ensemble Classifiers Integrated With LIME

In this era, when document preservation is of grave importance, the introduction of innovations to improve the quality of palm-leaf manuscripts becomes significant for the preservation of cultural heritage. These manuscripts deteriorate in various ways, which makes them difficult to preserve and read. The unique patterns of deterioration that can take place across each section of the same document are not easily captured by conventional methods. The current approach adopted uses chunks of palm-leaf manuscripts from documents and classifies them into deterioration types using multiple advanced machine-learning models. Multiple ensemble methods are used further to tune and increase the accuracy. After that, chunks are reassembled and tailored image enhancement is done based on the results. Wrapper forward feature selection to enhance the classification quality of the chunks and additionally Local Interpretable Model-agnostic Explanations (LIME) explainability is used to analyze features contributiing to the different classes. The proposed model outperforms traditional classification techniques by integrating chunk-wise classification with ensemble methods. It achieves a significant accuracy improvement, with Bagging Random Forest (RF) yielding the highest accuracy of 92%, compared to Support Vector Machines (SVM) and K-Nearest Neighbors (KNN) This demonstrates the model’s superior ability to handle the complexities of palm-leaf manuscript deterioration, providing a more accurate and reliable classification for document preservation.