“…Wang et al (2010Wang et al ( , 2012 decompose the hair segmentation into local parts and several other approaches (Wang et al, 2009(Wang et al, , 2011(Wang et al, , 2013 use region growing followed by refining regression on the coarse mask. Recent work (Chai et al, 2016;Qin et al, 2017;Guo and Aarabi, 2018;Levinshtein et al, 2017) based on FCNN models achieved good performance for practical applications. However, most of the methods only focus on constrained conditions such as head-shoulder images.…”