“…Results show that proposed method gets a Pr of 97.24%, an Re of 94.31%, and an F1 of 95.75%, which are higher than other state-of-theart pavement crack segmentation methods, such as Crack-Forest (Shi et al, 2016). Although the recall of proposed method is lower by a small margin than that achieved by (Zhao, Qin, & Wang, 2010), (d) local thresholding (Oliveira & Correia, 2009), (e) CrackForest (Shi et al, 2016), (f) method of Jenkins and colleagues (Jenkins, Carr, Insa-Iglesias, Buggy, & Morison, 2018), (g) method of Nguyen and colleagues (Nguyen, Le, Perry, & Nguyen, 2018), (h) method of Fan and colleagues (Fan et al, 2018), (i) proposed method Cheng, Xiong, Chen, Gu, and Li (2018), the precision of proposed method is higher than other methods, and the F1 score of proposed method outperforms other methods more than 3%. The high precision and F1 score show that the proposed method correctly segments more actual crack pixels and fewer fake crack pixels than other methods and the segmentation results of proposed method are closer to the ground truths.…”