Noise Presence Detection in QR Code Images

Abstract

A quick response (QR) code is symbols used to encode information such as key identifiers (website addresses, product, etc.) that can be printed and scanned electronically using image-based technology. However, it may include noise at the time of printing or scanning due to some environmental or mechanical factors. Therefore, the study analyzes various machine learning models to detect noise presence in QR code. For this, we first generated own dataset by creating 14,000 images of QR code, and then enhanced the dataset by adding several noises to the original QR code images. Later, it exploits several machine learning, deep learning and pre-trained models to segregate noisy images from original images. Experimental results show that ResNet101 and Xception models outperformed others by attaining 100% accuracy, recall, f1-score, and precision, each. Besides these, support vector machine (SVM) also performed better by accomplishing 99.6% accuracy on test set when trained over 70% of dataset.