To date, deep learning-based semantic segmentation techniques have been applied to crack segmentation problems; however, it requires much time and many resources to prepare a large volume of the ground truth of a dataset labeled at the pixel level. Hybrid crack segmentation (Kang et al., Autom Constr 118:103291, 2020) is based on the integration of a faster region-based convolutional neural network (faster R-CNN) as the deep learning-based object detection method and modified tubularity flow field (TuFF) as computer vision-based segmentation. In this paper, we further conducted experimental studies to investigate the performance of the developed hybrid concrete crack segmentation method using additional images with complex backgrounds in challenging environments.