.png){kind=icon}
.png){kind=icon}
Return
Gs-DeblurGANv2: a QR code deblurring algorithm based on lightweight network structure
.png){kind=icon}
PRE
AI
DOI：10.1007/s00530-024-01292-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Currently, QR codes are widely utilized in a variety of industries, including payment, shipping, and the industrial Internet of Things. However, during the detection and recognition process, QR code images are frequently impacted by external elements, including recording equipment, light, and filming angle, which causes fuzzy QR codes that cannot be read to provide accurate information. This research presents a quick deblurring technique (Gs-DeblurGANv2) based on lightweight networks to fix the potential blurring issue with QR images in real-world applications. The approach is based on the generative adversarial network concept, where the generative network employs the GhostNet lightweight module as the feature extraction network and introduces the feature pyramid structure, while the addition of the SKNet attention module optimizes the feature extraction from images. In addition, PatchGAN is used as the discriminative network and a dual-scale discriminator for global image and local features is set. Trained and tested under the QR code blurred dataset, the results show that Gs-DeblurGANv2 achieves 25.21 dB and 0.87 PSNR and SSIM for the deblurred images and the original HD images on the test set, and this result is better than previous research methods. The outcomes of the experiments demonstrate that the proposed Gs-DeblurGANv2 can efficiently make use of the feature information of QR code pictures and produce more effective deblurring performance.
Keywords:
QR code deblurring
Deep learning
Generative adversarial networks
Attention module
Two-scale discriminator
Journal
IF:
3.1
Papers:
2.7K
Citations:
2.7K
Organization
No organization information available