To overcome low efficiency of segmentation for multi-channel images when the color values of target region is similar to that of the background object, a multi-channel mean shift segmentation algorithm was proposed. The algorithm was achieved through mean shift algorithm for R, G, B three-channel hierarchical clustering in terms of the different sensitivity of R,G,B image to color. A reliability factor was introduced for separately evaluating the reliability of each pixel belonging to a cluster. Then the pixels of high reliability score were retained as the final segmentation results. Finally, a logical "OR" operation was used to fuse each single channel segmentation result to obtain the final result of image segmentation. The experimental results showed that compared to common segmentation algorithms, this algorithm could improve the segmentation results and improve the quality of the image segmentation. With the advantages of accuracy, the proposed algorithm was suitable for computer vision in defect detection of color printing and packaging image detection.