Method of Watershed with Threshold Decision for Multiscale Traffic Image Segmentation

The main problem that watershed image segmentation facing is over segmentation. The paper that combines adaptive threshold decision and watersheds improves the traditional image segmentation technology. Firstly, applying the wavelet transform to get the detail and approximation information. Based on the gradient magnitudes of the approximation image at the coarsest resolution, an adaptive threshold is used to remove minor gradient value. Finally, the watershed transform is applied. The result of experiments indicates the method produces a good segmentation and efficiently reduces over segmentation.