With respect to varied components and weak edge in tumor areas during the segmentation of computed tomography (CT) image of liver cancer ablation, a level set algorithm using the improved Chan-Vese model was proposed to accurately extract the contour of the hepatic tumor. According to significant difference of Gaussian mean and standard deviation between liver and tumor, the Gaussian mixture model was used to distinguish the subjection of pixels between target and background, and the bound terms of length and shape of edge gradient were combined to construct energy functions. The priori knowledge of tumor was applied to determining the initial profile of target, so that the active contour can converge on the edge of target area. By virtue of the verification algorithm of experimental data set of liver CT image, it is feasible to extract irregular contour of components such as inactivated or partially inactivated carcinoma tissues and iodized oil accumulation in liver. Experimental results showed that the average similarity value of our approach was higher than 0.87, the accuracy and precision of the improved algorithm were better than those of local Chan-Vese and local binary fitting models.