.png){kind=icon}
.png){kind=icon}
Return
Enhancing Digital Image Correlation Analysis for Poor-Quality Images Using Computer Vision Algorithms
.png){kind=icon}
PRE
AI
L
L
DOI：10.1007/s40799-026-00879-9.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Digital image correlation (DIC) is a non-contact, optical method for full-field deformation measurement. However DIC often fails to provide accurate full-field deformation results when applied to low-quality images. Therefore, a post-processing method of DIC based on computer vision algorithms for full-field deformation is proposed to address incomplete (non-full) displacement/strain field when measuring deformation in low-quality images. First, image noise are the main cause of inaccurate DIC results, and image anomaly detection is performed on artifact-contaminated images to identify and locate artifacts. The abnormal regions in the DIC deformation field are removed according to the detected artifact locations. Two strategies are investigated two approaches: one is to reconstruct the incomplete displacement field using Radial Basis Function (RBF) interpolation or Aggregated COntextual-Transformation Generative Adversarial Networks (AOT-GAN), subsequently calculating the full strain field; the other is to directly use RBF interpolation or AOT-GAN to reconstruct a complete strain field. The effectiveness of the proposed method is validated through comparative experiments and analysis on deformation measurements using oil sands images contaminated by artifacts underwater injection, involving three kinds of artifacts: scratches, lint and bubbles. The proposed method can not only evaluate the quality of experimental data and identify and locate the position of artifacts. Moreover, it can improve the DIC analysis results for digital images of poor quality, and obtain more accurate full-field deformation that meets the requirements of subsequent experimental analysis.
Keywords:
Image quality
Image anomaly detection
Interpolation
Generative adversarial networks
Digital image correlation
Post-processing method
Journal
E
IF:
1.9
Papers:
50
Citations:
1.6K
Organization
Cited Papers