In order to overcome the difficulties of description of texture components and coupling between layers of color image for color texture image segmentation,we proposed a combined non-local Mumford-Shah-MTV model. This model is under variational framework making use of the properties of MTV(Multi-channel Total Variation) regularizer in image coupling between layers and non-local operators in texture descriptions. Meanwhile, a binary label function is used to divide different regions in the model. In order to improve computational efficiency, we designed the ADMM(alternating direction method of multipliers) algorithm for the proposed model. The results of numerical experiments and qualitative and quantitative analysis demonstrate that the non-local Mumford-Shah-MTV model can obtain better characteristics for color texture image segmentation.