Color image interpolation is usually implemented by the componentwise interpolation scheme using gray image interpolation methods, which is insufficient and results in additional artifacts. In this paper, an effective color image interpolation method is proposed, which extends the classical bicubic convolution interpolation. Our improvement includes: considering the correlation between color channels; giving a new method of color edge detection and a novel scheme estimating the missing pixels. The proposed method can well preserve the edges and textures, and noticeably suppress common artifacts such as blurring, blocking and aliasing in interpolated images. The experimental results demonstrate that it outperforms significantly the bicubic interpolation in terms of both objective and subjective measures.