Research On Method Of Image Segmentation Based On Active Contour Model

The technology of image segmentation means extracting interested objects from image, and it is one of key issues in image processing. This dissertation presented some studies concentrated in five topics: image segmentation based on parametric active contour model (Snake model), image segmentation based on geometric active contour model, multi-object image segmentation, texture image segmentation, and application of developed curvature flow filter.Parametric active contour model incorporating regions information was studied for image segmentation. It is difficult for traditional (Gaussian, balloon force et al.) parametric active contour model to deal with automatic segmentation of weak edge medical image. Based upon analyzing balloon force model, a minimum variation Snake model is proposed and successfully applied to segmentation of weak edge medical image. In the model, variable force incorporating the information of foreground and background regions replaces constant force in the balloon Snake model. It makes curve to deform with the criterion of minimum variation of foreground and background regions. Experiments and results have proved that this proposed model is robust to initial contour placements, and it can segment automatically CT medical image of left Lateral Ventricle and Cerebellopontine angle. Due to its topology-adaptable ability,geometric active contour model is more suitable for complex image segmentation than traditional Snake model. A developed simple M-S model for image segmentation in geometric active contour model is presented based on intra-region similar and inter-region dissimilar properties. The model constructs an energy (cost) function, which is made of intra-region variations and weighting squares of subtraction of region mean values. Using gradient-descent methods, the energy function is minimized and we get a curve evolution equation that segments image. The developed simple M-S model is suitable for weak edge image segmentation by adjusting weighting coefficient. The experiments on tumor CT image segmentation show that geometric active contour model requires more computational load than Snake model but it can converiently control the topology construct of segmented image.To segment multi-object image, N-region segmentation problem is equivalent to N-1 two-region segmentation problems that are solved based on the developed simple M-S model. The experiments show that cytoplasm of immune cell is segmented out...