공간 가중치 마스크를 이용한 QR 코드 정보 추출 신경망 학습 및 데이터셋 개발 = Learning a Neural Network for QR Code Information Extraction Using Spatial Weighted Masks and Developing a Dataset

다국어 초록 (Multilingual Abstract)

In this paper, we propose a method to enhance the recognition and data extraction performance of QR codes projected onto screens using deep learning. The purpose of the developed technology is to convert screen image information encoded by QR code into voice in a classroom environment so that visually impaired people can take classes in real time. In order to increase the QR code recognition rate, custom learning data specialized for classrooms and screens were produced. We designed YOLO5n for QR code recognition and DeblurDiNAT for deblur, but proposed a spatial weight mask technique to improve information extraction performance from QR codes.The training data applied with the spatial weight mask showed 17% higher accuracy than other structures of the comparison target in the experimental environment.