Aiming at the oversegmentation problem that traditional segmentation methods are prone to affect by noise, a regionbased method that extracting adaptive coarse edge by using multidirection Gabor filters is proposed. Firstly, extract the multidirection magnitude responses by using imaginary part of Gabor filter and get the gradient magnitude at each pixel, then use the adaptive threshold method to extract coarse edge. Secondly, watershed transform on the coarse edge map to get the initial oversegmentation result. Finally, on the premise of not exceeding the region merging cost, the initial segmented regions are iteratively merged. The experimental results show that compared with Canny, region growing and LOG methods, the proposed method relieves the oversegmentation problem effectively and the average computational complexity has been reduced by about 21.5%.