A watershed-based image segmentation using effective region merging strategy. The proposed algorithm is a hybrid segmentation technique. Firstly, a filter is implemented to detect the boundary of the objects in the input gray-scale image and we mark the minimum gray value of pixels before adopting watershed transformation. Each region is labeled by a unique number after the transform. Then each region is represented by a graph node and the neighboring nodes of each node are saved in a matrix for the computation of region dissimilarity between the adjacent nodes according to three features: intensity mean, intensity variance and the number of pixels in a region. Last, the two regions which have minimum cost are merged while the information of adjacent nodes' relationship, cost among the merging nodes and their neighboring nodes and the labels of the regions are updated. This will be iterated until the final segmentation result. The experimental result shows that comparing to normalized cut algorithm the proposed method is more efficient and achieves more reasonable segmentation.