Traditional transition region extraction methods are based on gradient operator. They are sensitive to noise and restricted by L_(low) and L_(high) . By analyzing properties of transition regions, a novel local complexity based on transition region extraction method (C-TREM) was presented. C-TREM is a direct method to extract transition regions. The filtering ability of local complexity improves the ability of C-TREM to deal with noises. C-TREM depends no more on L_(eow) and L_(high). Experimental results demonstrate that C-TREM significantly outperforms the conventional gradient-based transition region extraction methods (G-TREM)