Multilingual Receipt Image Preprocessing Optimization for OCR

Data extracted from receipts are necessary for various applications in many industries, and this extraction process can be significantly simplified through deep learning. However, high-quality receipt images are essential for algorithms to produce accurate models. This paper focuses on the techniques to improve the image preprocessing process and the information extraction methods applied in Vietnamese and English. The paper will provide an overview of the research question and scope of the review, followed by a literature review of related works. The paper then describes the image preprocessing techniques used in the research, including measuring image blur using the Fuzzy C-Means clustering and removing the background through thresholding techniques such as brightness and chromaticity distortion. Finally, the paper discusses the information extraction methods employed in Vietnamese and English, including using two specialized models. ViptOCR and Convolutional Neural Network.