Speckle Texture Quality Guided Digital Image Correlation for Measurement under Disturbance Conditions

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract In practical digital image correlation measurements, speckle degradation and loss of effective information often occur due to various disturbances. Despite its widespread application in displacement measurement without physical contact, these factors can significantly affect measurement accuracy. Considering the influence of speckle patterns on digital image correlation, a speckle texture quality factor is proposed, which evaluates speckle texture quality and optimizes the correlation coefficient formulation. A new seed point selection strategy is designed, and the optimized correlation coefficient is used as the propagation criterion in integer pixel displacement estimation. In the subpixel stage, a Gaussian weighting function is further incorporated to enhance displacement estimation accuracy. Through these improvements, the utilization of reliable speckle regions is strengthened, thereby reducing the influence of disturbances on measurement accuracy. The proposed method was evaluated using sinusoidal image sequences generated from real speckle images under four representative disturbance conditions, namely non-uniform illumination, low contrast, local speckle loss, and additive Gaussian noise. Ablation experiments confirmed the individual contributions of the speckle texture quality factor, seed point selection with propagation-based integer-pixel estimation, and Gaussian-weighted subpixel fitting. Comparative experiments with three comparison algorithms showed that the proposed method achieved lower RMSE under illumination variation, speckle loss, and additive Gaussian noise, and remained competitive under low contrast conditions. The SNR results further demonstrated improved measurement stability under continuously varying disturbance intensity. The influence of Gaussian width was also examined, and moderate values generally provided stable performance. Finally, translation stage experiments under different illumination conditions verified the practical applicability of the proposed method. For the 12 mm and 75 mm lens image sets, the proposed method achieved average RMSE values of 0.1534 pxl and 0.5141 pxl, respectively, outperforming the comparison algorithms.