Abstract: A new watershed algorithm based on multi-scale and self-adaptive marker-extraction is proposed to overcome over-segmentation.Firstly,multiple scale filtering is executed on the V component of HSV space,and image after filtering is transformed from HSV space to RGB space.And then,multiple scale filtering is executed on image gradient.Finally,H-minima transformation is used to adaptively extract markers.At last,the watershed transformation of modified gradient is performed.The experimental results show that the proposed algorithm is obviously effective for the over-segmentation,and obtain meaningful and more reasonable regions.