Effective image segmentation is a challenging task in computer vision and object recognition. Watershed segmentation algorithm is simple and accurate, but over segmentation is a major drawback of this method. An improved watershed image segmentation algorithm is proposed to solve the problem of over-segmentation by classical watershed algorithm by preprocessing step to limit the number of regions. Then detecting edges of image and computing watershed transformation is done. The experiment results show that this method is effective on solving the problem of over-segmentation.