HMPLMD: Handwritten Malayalam palm leaf manuscript dataset

Area of Science:

• Digital Humanities
• Computer Vision
• Document Image Analysis

Background:

• Degraded document recognition, especially for historical manuscripts, heavily relies on effective document enhancement techniques.
• Deep learning models show significant promise for document enhancement compared to traditional methods, but require extensive ground truth data.

Purpose of the Study:

• To address the challenge of limited ground truth data for training deep learning models for degraded palm leaf manuscript analysis.
• To propose a novel dataset and its corresponding ground truth for Handwritten Malayalam Palm Leaf Manuscript Dataset (HMPLMD) to advance palm leaf image analysis.

Main Methods:

• Development of a new dataset, HMPLMD, focusing on degraded palm leaf manuscripts.
• Creation of corresponding ground truth data for the proposed dataset.
• Utilizing palm leaf manuscripts from Kambaramayanam and Jathakas for experimental validation.

Main Results:

• The proposed dataset and ground truth samples are crucial for developing specialized deep/transfer learning models.
• The dataset aims to tackle challenges in binarization for ancient, degraded documents.

Conclusions:

• The HMPLMD dataset is a significant contribution towards improving the performance of deep learning models in recognizing degraded historical documents.
• This resource will facilitate advancements in the automated analysis and preservation of palm leaf manuscripts.