The Gradient-Domain fusion is an effective method for image compositing, but suffers from color distortion when the hue of target image is quietly different from that of source image. The distortion can hardly be eliminated by boundary optimization. This paper proposes a method to correct the artifact of color distortion induced by gradient-domain fusion while keeping the seamless boundary. First, Poisson cloning is applied to luminance component for keeping the local contrast of intensity. Then, for each pixel of source image, a color belief is used to guide color correction and is estimated based on general geodesic distance transform for source image. Next, the color components of each pixel are corrected by minimizing an object function with corresponding color belief, while the object function takes neighborhood colors propagation, the original colors obtained by Poisson cloning are also taken into consideration. The final composites are obtained simply by combining the corrected color and the cloned luminance. Experimental results demonstrate that the proposed method reduces the color distortion of composition effectively while keeping the seamless boundary and requires only few interactions from user. Compared to other gradient-domain based image compositing methods that only optimize boundary conditions, the proposed method keeps both the original color and the relative changes of colors. The use of a soft color belief to compositing provides a smooth transition from foreground to background, which can not be done by hard constraints.