Handwritten Form Recognition Using Artificial Neural Network

Form Recognizer aims to build a deep learning model to extract handwritten text from a scanned Permanent Account Number (PAN) application form and convert them into digital format or editable text and store it in an excel file for further processing like statistical analysis or machine learning. The Learning model is based on the Convolution Neural Network (CNN) for the feature extraction and higher end classification. To accomplish this task, the handwritten forms are scanned, preprocessed to remove noise and handwritten fields are extracted. OpenCV is used to get the contours of the characters in the extracted images. This approach gives better accuracy than using plain CNN without out contours. The CNN model gives an accuracy of 91% on merger of numbers, uppercase and lower-case alphabets of EMINST dataset. Further, handwritten form recognizer system is built by incorporating this learning model, which is in turn integrated with preprocessing and segmentation methods. Finally, the output of the system is stored in a CSV file.