Study On Object Segmenting Technology For Dense Grain Image

In the experimental mechanics, there is an enormous significance on studying the motion of sand grains at high pressures. Tracking grains in the case of pressure and getting the relative position in process of movement are the base of mechanical analysis. For getting the resolution of tracking the dense grains images, this paper mainly study the target grains segmentation in multicolor grains image which based on color model, and target grains segmentation in gray grains image based on boundary detection and Level Set, et al.At first, this paper studies the theoretic method and research actuality in domestic and overseas, shows some basic theories and methods as the image pretreatment, transformation of multicolor space, Mathematical Morphologic methods, boundary detection, active contour technology. As well as there are some introduction of the application of these technologies in the study of this paper.There exist many differences between the color of background grains and the color of grains to be extracted. For making full use of the color information, the paper transforms the multicolor image to HSV space. By the study of the target grains and the background grains, we can get the Gauss Model of the samples. Based on this model, we can get the degree of similarity by calculating the pixels of the whole image to the target grains and background grains one by one, and accumulates them to a class which has a high degree of similarity. The binary effect will shows in the multicolor, by this we can divide the target grains and background grains into two parts. Statistics show that the results of tests to identify the target grains in the accuracy of 90%.For the complex dense gray grains images, the paper gives a method based on boundary detection and Level Set repairing. First, we use the improved adaptive canny algorithm to do the edge detection, then get rid of the boundary noises by using the mathematical morphologic methods, and improving the quality of boundary information; Second, selecting seed points, according to the sub-image histogram concentration automatically select seed points; the detecting rays are emitted from each seed to the around to detect the boundary positions, record the boundary points. By the judge Mechanism we can find the pseudo boundary points which is created by the deformity of boundary, the overlap grains or noises, and adjust the pseudo boundary points by the grains' nature of analogous oval, and the get the whole contour of the grains area; Finally, by using the Level Set method, the grains area which get in above process can be fitted to make the segmentation effect similar to the real shape. Compare with the results of watershed, regional growth, the segmentation results of the method which was proposed in this paper with the gray grains are closer to the true shape of the objects, and have a better effect in division.