Reconstructing unreadable QR codes: a deep learning based super resolution strategy

Area of Science:

• Computer Vision
• Artificial Intelligence
• Digital Imaging

Background:

• Quick Response (QR) codes are vital for digital transformation, enabling fast information sharing.
• Scanner limitations cause distortions like low resolution and misalignment, hindering QR code readability and data extraction.
• These issues can increase processing times and introduce security risks.

Purpose of the Study:

• To evaluate the effectiveness of four super-resolution models in improving the readability of distorted QR codes.
• To mitigate issues such as low resolution, rotation errors, and misalignment in QR code scanning.
• To assess the performance of super-resolution models on both simulated and real-world degraded QR codes.

Main Methods:

• Utilized four super-resolution models: Enhanced Deep Super Resolution (EDSR), Very Deep Super Resolution (VDSR), Efficient Sub-Pixel Convolutional Network (ESPCN), and Super Resolution Convolutional Neural Network (SRCNN).
• Created a dataset of 16,000 computer-generated QR codes with simulated scanner-induced distortions.
• Applied super-resolution models to 4,593 unreadable real-world QR codes and a separate set of 2,899 simulated unreadable QR codes.

Main Results:

• EDSR, VDSR, ESPCN, and SRCNN successfully recovered 4,261, 4,229, 4,255, and 4,042 of the unreadable real-world QR codes, respectively.
• When trained with OpenCV's deep learning detector, these models successfully read 2,891, 2,884, 2,433, and 2,560 of the simulated unreadable QR codes.
• The study demonstrates significant improvements in QR code readability using super-resolution techniques.

Conclusions:

• Super-resolution models are highly effective in enhancing the readability of degraded or low-resolution QR codes.
• These models offer a viable solution to overcome common scanning limitations and improve data extraction reliability.
• The findings support the integration of super-resolution techniques for more robust QR code applications in digital systems.