Automated and accurate initialization of digital image correlation for large deformation measurement

Digital image correlation matches the corresponding locations in the reference and deformed images by optimizing the correlation of the related intensities. Iterative algorithm is regarded as the most effective approach to solving the optimization, but it requires accurate initial guess of the deformation parameters to converge correctly and rapidly. This paper presents a fully automated method which provides accurate initialization for all points of interest in a deformed images and deals with large rotation and heterogeneous deformation. Image features are extracted and pre-matched in the reference and the deformed images. The deformation parameter of a sample point is initialized by the mapping function fitted to the matched features in the vicinity. Once the subsequent iterative optimization achieves a qualified correlation measure, the optimized parameter is used to initiate a parameter transfer. To account for potential deformation difference between successive points, propagation functions are used during the transfer, which are analytically derived to accurately relate the parameters of two points separated by a given distance. The parameter transfer is conducted by the quality of correlation optimization and continued till all the points have been analyzed. Results on both simulated deformations and real-world experiments demonstrate that image features can be reliably matched and automatically generate qualified seed points even in the presence of complex transformation. Parameter transfer using propagation function enables rapid and correct convergence of the nonlinear iterative optimization, allows more flexible choice on the interval between adjacent sample points, and meanwhile handles the heterogeneity of the deformation field.