A feature-level image fusion scheme based on segmentation and iterative fashion is introduced in this paper. In this scheme, a total variation based energy model is proposed to estimate the contribution of each feature on the source images, the features are obtained based on segmentation of the source image. A larger energy indicates the feature with a strong pattern on its associated region. Furthermore fusion strategies may be performed by measuring the contributions of the associated features on the images. Different from previous methods, measurement of the contributions of the features in proposed are performed when the features has been fused. Therefore a global optimization could be achieved. In order to keep a low computational cost, the orders of regions have not been considered in current research. Even so, better results have been obtained on several test images, and the performance assessment of the final fusion results also are evaluated by using several classical evaluation methods like Root Mean Square Error and Peak Signal to Noise Ratio.