This paper presents a scheme for color image segmentation based on watershed algorithm that solves over-segmentation problem. A lexicographical order is adopted on HSI (Hue-Saturation-Intensity) color model so that morphological operations can be applied on color images directly. A marker image is then defined using both chromaticity and intensity information of the image. Together with multi-scale gradient image and marker image, watershed segmentation can make sure to partition image into meaningful objects and avoid further segmentation of homogeneous regions. The results of the proposed algorithm are compared with those of other standard methods. Preliminary experiments have shown a better result and indeed solved over-segmentation problem.