Methods for threshold optimization for images collected from contrast enhanced concrete surfaces for air-void system characterization

Several automated procedures for the characterization of the air-void system of hardened concrete rely on a contrast enhancement step to make air-voids appear white and aggregates and paste appear black. Pixels in the digital image darker than a selected threshold level are classified as non-air, pixels brighter are classified as air. Laboratories that perform air-void testing typically have a large number of samples with corresponding results from manual operators. Proponents of automated methods often take advantage of this fact by analyzing the same samples and then comparing results. A similar iterative approach is described here where scanned images collected from a significant number of samples are analyzed and the threshold optimized to best approximate the results of the manual operator. The results of this calibration procedure are compared to an alternative approach based on more rigorous digital image accuracy assessment methods employed by the remote sensing/satellite imaging community.