Efficient Data Extraction from Handwritten Forms: A Structured Pipeline Solution
Shivani Anil Patel, Krisha Borana, Neha Katre, Vinaya Sawant, Harshal Dalvi · 2024
This paper presents an approach to systematically extract data from handwritten forms and convert them to structured data. Unlike traditional handwriting character recognition systems, proposed approach integrates techniques such as alpha blending for precise image alignment and deep learning-based models for enhanced text recognition accuracy. The pipeline begins with the automated detection and delineation of form fields (like Name, DOB, Occupation) followed by sophisticated pre-processing steps that include noise reduction and threshold management to ensure optimal image quality. The system then employs the proposed trained model to recognize handwritten text and match it to the extracted entities like key value pairs. The values are then appended to structured data. The proposed method has been evaluated, demonstrating accuracy and reliability in diverse real-world scenarios. This method would bypass the need to individually extract values and handwritten texts, saving time and effort.