Abstract: In order to overcome the impact of mismatched feature points and uncertainty in matching accuracy on initial guess of small deformation based on scale-invariant feature transform (SIFT) feature matching, least absolute deviation algorithm and Hough transform algorithm which are reasonably resistant to outliers caused by mismatches were introduced to obtain initial deformation values. The reliability of initial guess of small deformation calculated from the least absolute deviation algorithm and Hough transform algorithm were verified with the simulated speckle pattern experiments, and the results of two algorithms were compared with that of random sample consensus algorithm. In addition, the reliability of the initial guess algorithm in a real scene was verified with a similar simulation experiment. The results show that the standard deviations of the initial displacements of the absolute deviation algorithm are ranging from 0.0033 pixel to 0.0068 pixel, lower than that of the Hough transform algorithm and the random sampling consistency algorithm in the case of small displacement. The average number of iterations for the correlation search using its initial displacement is 3.692 to 4.370 in the digital image correlation method. For the damaged region in the speckle pattern during the experiment, there is a significant difference between the initial displacement of the least squares algorithm and the displacement measurement of the digital image correlation method, with a maximum of 11.80 pixel and a minimum of −7.35 pixel. The least absolute deviation algorithm still provides reliable initial values, but the digital image correlation method has a risk of deformation measurement failure. This study provides some reference for initial guess of small deformation in the digital image correlation method.