Multi-scale image segmentation based on morphology

The combination of scale-space and water-shed is a common way to implement multi-scale segmentation; however, classical scale-space often surfers from the essential contours lost, low localization precision and noise caused spurious contours. Though watershed transform can produce closed contours, it tends to over-segmentation. A method for multi-scale image segmentation using morphological scale-space and modified gradient watershed is proposed to improve the above performance. First, morphological reconstructive opening and closing scale-space is used to simplify the original image, and then on the basis of the simplified images, modified gradient watershed transform is employed to detect region contours. During the image simplifying process, noise and small details are removed while essential contours preserved owing to the strong filtering and contour preserving properties of the morphological reconstructive opening and closing scale-space. Modified gradient avoids too many irrelevant minima to be detected, and over-segmentation is eliminated. Simulations show that this method can efficiently not only segment image without over-segmentation, but also satisfy the scale monotonicity of scale-space.