As existing objective evaluation index for color image fusion is inconsistent with the human vision perception, a non-reference index based on quaternion convolution was proposed. First, a color image was modeled in a holistic manner, in which the color information of the color image was considered fully as a whole. Then, the quaternion-valued edge detection template and the color image were used to do a convolution operation and to get the detailed color information. Furthermore, the image definition and useful information from the fusion were measured and they were given by weight modes. Finally, a set of quantitative computations for the fusion images were performed and objective evaluation results were given. The experimental results show that the proposed method can utilize the color information and other detail information obtained by human vision. It works better than the traditional methods, and shows a better stability in the color image fusion evaluation. The evaluation results of proposed method are consistent with the human vision perception, and fulfill the needs of objective color image fusion.