A Layout Independent Deep Learning Framework for Recognition of Courtesy-Amount in Bank-Cheque Image

This article proposes a deep learning-based end-to-end framework to recognize the courtesy amount from an Indian bank cheque without any detailed layout specific analysis of the cheque template. The proposed framework employs a refined Mask-RCNN-based segmentation module to localize the courtesy amount in the bank cheque and subsequently, extract the digits of the segmented courtesy field region. The sequence of the extracted digits is further processed by a CNN-based digit recognition module to recognize the courtesy amount. The proposed model is evaluated using standard IDRBT Cheque Image Dataset and the experimental results are found to be promising in terms of localizing the courtesy field and recognizing the respective handwritten amount.