One main limitation of conventional digital image correlation (DIC) methods is their incapacity to treat with large, discontinuous displacements. This paper seeks to resolve this issue using a variational method. Specifically, the objective function is written in a form of L¹‐norm instead of L²‐norm for discontinuity preservation. This convex, non‐differentiable objective function is resolved by a proximal gradient method. Then, the entire correlation process is embedded into a multi‐level image pyramid, enabling large displacements to be measured accurately. The proposed method is first validated on images with prescribed small displacements; its performance to treat with discontinuous, large displacements is then examined. The results show that the proposed method exhibits an acceptable performance when measuring some extreme displacements up to 16.5 pixels in a 500 × 500 pixels resolution image. Concerning subpixel accuracy, the errors can be lessened to a very low level with optimised parameters even for a dense correlation at each pixel. Good performance is also demonstrated in forms of noise proof and illumination fluctuation tolerance. Finally, an application is conducted on an experiment concerned with cracking in desiccated soils, in which the proposed method succeeds in tracking the nucleation and propagation processes of multiple cracks.

中文翻译：

---

从数字图像测量大的不连续位移

常规数字图像相关（DIC）方法的一个主要限制是它们无法处理大的，不连续的位移。本文试图使用变分方法来解决此问题。具体来说，目标函数以L ¹范数而不是L ²的形式编写-规范以保留不连续性。这个凸的，不可微的目标函数通过近端梯度法解决。然后，将整个相关过程嵌入到多级图像金字塔中，从而可以精确地测量大位移。该方法首先在指定位移小的图像上得到验证。然后检查其在不连续大位移情况下的性能。结果表明，当在500×500像素分辨率的图像中测量一些最大位移为16.5像素的极端位移时，该方法表现出可接受的性能。关于子像素精度，即使对于每个像素处的密集相关性，也可以通过优化参数将误差降低到非常低的水平。还以防噪和照明波动容忍的形式证明了良好的性能。最后，在与干燥土壤开裂有关的实验中进行了应用，该方法成功地跟踪了多个裂纹的成核和扩展过程。