Using Image Gradients to Improve Robustness of Digital Image Correlation to Non-uniform Illumination: Effects of Weighting and Normalization Choices
KAUST DepartmentComposite and Heterogeneous Material Analysis and Simulation Laboratory (COHMAS)
Mechanical Engineering Program
Physical Science and Engineering (PSE) Division
Online Publication Date2015-03-05
Print Publication Date2015-06
Permanent link to this recordhttp://hdl.handle.net/10754/566187
MetadataShow full item record
AbstractChanges in the light condition affect the solution of intensity-based digital image correlation algorithms. One natural way to decrease the influence of illumination is to consider the gradients of the image rather than the image itself when building the objective function. In this work, a weighted normalized gradient-based algorithm, is proposed. This algorithm optimizes the sum-of-squared difference between the weighted normalized gradients of the reference and deformed images. Due to the lower sensitivity of the gradient to the illumination variation, this algorithm is more robust and accurate than the intensity-based algorithm in case of illumination variations. Yet, it comes with a higher sensitivity to noise that can be mitigated by designing the relevant weighting and normalization of the image gradient. Numerical results demonstrate that the proposed algorithm gives better results in case of linear/non-linear space-based and non-linear gray value-based illumination variation. The proposed algorithm still performs better than the intensity-based algorithm in case of illumination variations and noisy data provided the images are pre-smoothed with a Gaussian low-pass filter in numerical and experimental examples.
CitationXu, J., Moussawi, A., Gras, R., & Lubineau, G. (2015). Using Image Gradients to Improve Robustness of Digital Image Correlation to Non-uniform Illumination: Effects of Weighting and Normalization Choices. Experimental Mechanics, 55(5), 963–979. doi:10.1007/s11340-015-9996-1
SponsorsFunding for this research was provided by KAUST baseline funding. The authors are grateful to KAUST for its financial support. We are also grateful to Justin Blaber of the Georgia Institute of Technology for providing open source codes .