Subpixel microscopic deformation analysis using correlation and artificial neural networks

Abstract

Microscopic deformation analysis has been performed using digital image correlation and artificial neural networks (ANNs). Cross-correlations of small image regions before and after deformation contain a peak, the position of which indicates the displacement to pixel accuracy. Subpixel resolution has been achieved here by nonintegral pixel shifting and by training ANNs to estimate the fractional part of the displacement. Results from displaced and thermally stressed microelectronic devices indicate these techniques can achieve comparable accuracies to other subpixel techniques and that the use of ANNs can facilitate very fast analysis without knowledge of the analytical form of the image correlation function.