Research On Image Segmentation Based On Morphological Gradient And Watershed

Image segmentation is the key step of the transformation from the image processing to image analysis. And it is also a basic technique in computer vision. This is because that the technique such as image processing, objects separating, feature extracting, parameters measuring can transfer the original image to a more compact form, making the high level image analysis and understanding possible. Therefore, the image segmentation attracts the regard of many people, and variety algorithms come out. Among all the algorithms, watershed algorithm is popularly used because of the characters of fine segmentation result, easy to implement etc.The draw back of the watershed algorithm is that it is vulnerable to the noise and quantity error in the image and will divide the image to trivial areas. This is also known as over-segmentation. In order to solve the problem, the common method is to combine the areas after over-segmentation reducing the impact of the trivial areas. An algorithm is presented intending to make pre-processing before the watershed in order to decrease or abstain the over-segmentation.To achieve this goal, the mathematic morphological theory is introduced. The morphological filter is used to filter the noise in original image and then the morphological multi-scale gradient operator is used to calculate the gradient of the image. Based on the theory that any area of watershed segmentation is directly caused by a gradient valley in the image, a valley-filling method is proposed. The amount of the regions in the result of watershed is controlled by adjusting a threshold. On the other hand, the edge information of the original color image is detected by the classic Canny operator, and is used to enhance and revise the edge in gradient image. As a result of this, the contours of regions will perfectly coincide with the edge of targets in the image while filling the valleys.Simulations show that the proposed method has an impressive effect on over-segmentation, and the feature of degree-controllable should have its value in applying different requirements.