Super-Resolution of Defocus Thread Image Based on Cycle Generative Adversarial Networks - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Super-Resolution of Defocus Thread Image Based on Cycle Generative Adversarial Networks

Résumé

The dual camera calibration measurement method can realize low-cost and high-precision bolt dimension measurement by using two microscope cameras. But the height difference between the thread crest and root exceeds the depth of field, and the thread image becomes defocus, which seriously affects the measurement accuracy. For this reason, a super-resolution method for defocus thread image based on cyclic generative adversarial networks is proposed. We collected focus thread images and defocus thread images as training data. Two encoders are used in the generation network to extract image defocus features and content features. And a sub-pixel convolution layer is added to the decoder to achieve image super-resolution. A loss function based on adversarial loss and cycle-consistent loss is constructed to realize unsupervised training of the network, thereby achieve super-resolution of defocus thread images. The experimental results show that, in the simulated defocus images, the method has superiority in image detail preservation, sharpness improvement and peak signal to noise ratio. In the bolt dimension measurement task, it can effectively reconstruct the clear thread image and thus provide the measurement accuracy to 0.01 mm.
Fichier sous embargo
Fichier sous embargo
0 5 14
Année Mois Jours
Avant la publication
mercredi 1 janvier 2025
Fichier sous embargo
mercredi 1 janvier 2025
Connectez-vous pour demander l'accès au fichier

Dates et versions

hal-04178724 , version 1 (08-08-2023)

Licence

Identifiants

Citer

Pengfei Jiang, Wangqing Xu, Jinping Li. Super-Resolution of Defocus Thread Image Based on Cycle Generative Adversarial Networks. 12th International Conference on Intelligent Information Processing (IIP), May 2022, Qingdao, China. pp.457-472, ⟨10.1007/978-3-031-03948-5_37⟩. ⟨hal-04178724⟩
12 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More