Abstract: An improved marker-controlled watershed segmentation method is proposed to reduce the oversegmentation of watershed algorithm.Firstly,regional minimas in the gradient image are extracted according to the connectivity of adjacent pixels.Then,a maximum entropy threshold method is used to exclude the local minimas generated by the noise and texture.Finally,regarding the modified minimum points as markers,the watershed algorithm is applied to the modified gradients by the markers.The advantages of this method is that makers are extracted adaptively without the need for prior knowledges,which overcomes marker extraction difficulties.Experimental results show that the proposed algorithm can be effective overcomes the over-segmentation problem of watershed.