The statistical image modeling method uses some distributed model of parameter control to describe texture and its characteristics, and parameter estimation is the crucial issue of the method. In this paper, a novel texture feature extraction method is proposed, which adopts statistical image modeling with generalized Gamma distribution and generalized Von Mises distribution to extract texture features through logarithmic cumulants based parameter estimation in the dual-tree complex wavelet transform domain. Experimental results on VisTex and Brodatz databases show that the proposed method can effectively capture texture features of image, and achieve higher classification accuracy rate.