Automated Receipt Scanning Using Convolutional Recurrent Neural Network (CRNN)

Automated receipt scanning leverages optical character recognition (OCR) to transform printed or digital receipts into structured digital formats. This OCR technology involves scanning images of receipts and recognizing text on the receipts. The challenge relies on the process of recognizing the text although there are varieties of OCR libraries available. One of the main challenges is the variation of text on the receipt that needs to be recognized by the OCR. Hence, the OCR is required to be robust and capable of handling the various variations in the text. On the other hand, deep learning-based techniques are widely adopted in performing automatic recognition tasks including in security, finance, and education fields. Thus, this study aims to investigate the performance of a deep learning-based technique, Convolutional Recurrent Neural Network (CRNN) in performing OCR tasks. This study is concentrating on the implementation of CRNN in the OCR. Accuracy test and Character Error Rate (CER) test were executed to measure the performance of CRNN for OCR in performing automatic receipt scanning. Based on the 30 test images, the accuracy rate achieved is 66.67% and the CER is 10.48%. This result could be further improved by including a preprocessing stage that could help to preprocess images into improved images.