In this paper, an adaptive edge preserving variational framework for color image regularization is proposed. Unlike previous methods that emphasize only on image regularization term or use constant image fidelity term, in this framework, the balance between image regularization term and fidelity term is elegantly reached by adaptively assigning edge regions higher fidelity weight than homogenous regions. The adaptive fidelity weight is decided by the proposed edge indicator function together with the estimated noise variance. The proposed approach achieves both good noise removal and edge preserving. Experimental results show the improved performance of our framework compared to other existing methods even in highly noised conditions.