Image Classification of Transfer Slips and Optical Character Recognition

Abstract

This article presents the classification of transfer slip images and Optical Character Recognition (OCR) use of automatic learning techniques that mimic the human neural network. The approach utilizes a Convolutional Neural Network (CNN) algorithm specifically designed for classifying transfer slip images, comprising three 2D convolutional layers combined with Max Pooling to prevent overfitting during the classification process. Subsequently, image segmentation is performed using YOLOv5 to identify the bank names, followed by OCR using Tesseract OCR technology to read the information on the transfer slips. The dataset was randomly split into a training set (80%) and test set (20%). Within the training set, 20% of the data was used as a validation set. The model was trained using a batch size of 32 and 50 epochs, achieving a classification accuracy of 99%. After classification, the segmented images were used to identify the bank names, and once identified, the images were subjected to OCR to extract the text. The results were displayed in a human-readable format and exported in JSON format.