多分辨率下自适应阈值判定的分水岭车辆图像分割算法

Method of Adaptive Threshold Decision and Watersheds for Multiscale Image Segmentation

针对分水岭算法产生过分割现象,提出一种基于自适应阈值判定分水岭算法的图像分割方法,首先对图像进行多分辨率小波分解,再在某一尺度下计算梯度图像,并对该梯度图像进行自适应阈值判别,最后使用分水岭算法进行图像分割。实验结果表明该方法能取得较好的分割效果,有效的降低了过分割现象。

The problem of watershed transform is over-segmentation. This paper proposed a segmentation technique that combined adaptive threshold decision and watersheds for multiscale image segmentation.The wavelet transform then was applied to the gray image,producing detail and approximation coefficients.Gradient magnitudes of the approximation image at the coarsest resolution were computed,and an adaptive threshold was used to remove small gradient magnitudes.The watershed transform was then applied.The result indicated that this method could generate a good segmentation for gray images and efficiently reduce over-segmentation.