Image Segmentation Using a Hybrid Gradient Based Watershed Transform

A watershed transform segmentation method based on hybrid gradient that combines intensity and texture visual cues is proposed. Firstly a bilateral filtering method derived from robust statistics is used to extract the intensity gradient. Secondly a Gabor filter bank is applied to extract texture features. With a smoothing post process, the texture gradient is extracted. Then by morphological dilation and normalization process texture and intensity gradients are fused to form the hybrid gradient. At last the marked watershed transform on the hybrid gradient image is carried out to segment the image. The experiment results show that the proposed method is effective in generating accurate primitive-objects boundaries and meanwhile reducing the over segmentation of image.