Fast subpixel digital image correlation using artificial neural networks

Digital image correlation has been used to measure microscopic deformation in thermally stressed microelectronics devices. Displacement precisions of better than 0.03 pixels have been achieved by combining nonintegral pixel shifting of subimages and artificial neural networks (ANNs). The ANNs are trained to estimate the subpixel element of the object displacement from the digital correlation. Although similar accuracies can be obtained by curve-fitting to the correlation peaks and differentiating, the neural approach has the advantage that it allows fast subpixel. displacement analysis over a range of object textures without knowledge of the analytical form of the correlation peaks