Research On Watershed Segmentation Algorithm Based On Multi - Feature Marker

As the resolution of remote sensing images improves, the ground objects in the images become abundant and complex. The difference between ground objects sizes and between the pixels features of the same ground object becomes larger which challenges the traditional image segmentation a lot. As an effective method of image segmentation, the watershed transformation has become popular on application and research since it was introduced in image segmentation. The watershed transformation is essentially a kind of region growing method which is usually applied on the grad feature images. Influenced by the image noise and the density texture features, the images are usually heavily over-segmented on the direct watershed transformation. The key points of watershed transformation extraction of grad features of the image and the extraction of the marker image. To improve the over-segmentation result of watershed transformation, this paper studies on the extraction methods of the grad feature images and the marker images. On the aspect of grad feature extraction, this paper analyzes all the methods of grad feature extraction in the spatial domain. As the traditional extraction methods of the grad images is sensible to the image noise and the changes of the environment, this paper studies on the Phase Coincidence to extract edge features of the images in frequency domain to improve the situation. On the aspect of the marker images extraction, this paper researches on the extended minimum method, the threshold segmentation of grad feature image and texture feature energy image to get the marker images. And this paper also studies on the decision of the threshold value and provides a method to set the proper threshold which is proved to be available and adaptive by experiments. After marking the minimum values of the grad feature images by the marker images, the over-segmentation of the Flood Vincent Watershed Transform was improved effectively.