This paper describes a novel multiphase level set active contour segmentation method enabling simultaneous segmentation of multiple complex objects. The proposed technique supports preservation of prior topological co...
详细信息
ISBN:
(纸本)9781424456536
This paper describes a novel multiphase level set active contour segmentation method enabling simultaneous segmentation of multiple complex objects. The proposed technique supports preservation of prior topological configuration between different objects as well as to provide soft shape constraints. Whereas each contour evolves using its own evolution equation the contours are constrained and interrelated to each other by specially designed normal velocity component in the contours evolution equations. This velocity component is calculated using adaptively estimated set of medial axes between the corresponding contour and all other contours. In the proposed implementation the medial axes are represented by level set functions calculated from the level set functions representing active contours. By design, the proposed constraint can be used with edge, area or hybrid based methods and can easily incorporate additional constraints. The paper shows performance of the method on simulated as well as real images.
Disparity contours arc easily computed from stereo image pairs, given a known background geometry. They facilitate the segmentation and depth calculation of multiple foreground objects even in the presence of changing...
详细信息
ISBN:
(纸本)9783540896814
Disparity contours arc easily computed from stereo image pairs, given a known background geometry. They facilitate the segmentation and depth calculation of multiple foreground objects even in the presence of changing lighting, complex shadows and projected video background. Not relying on stereo reconstruction or prior knowledge of foreground objects, a disparity contour based image segmentation method is fast enough for some real-time applications on commodity hardware. Experimental results demonstrate its ability to extract object contour from a complex scene and distinguish multiple objects by estimated depth even when they are partially occluded.
The methods proposed in this paper improved the classical pulse coupled neural network (PCNN). Through adding the extracted edges of the objects into the classical PCNN, the synchronous bursts of non-linking neurons w...
详细信息
The methods proposed in this paper improved the classical pulse coupled neural network (PCNN). Through adding the extracted edges of the objects into the classical PCNN, the synchronous bursts of non-linking neurons with different input were generated in the proposed PCNN model in order to realize the multi-object segmentation. The paper provided the criterion of choosing the dominant parameter (the linking strength β) automatically, which determines the synchronous-burst stimulus range. At the same time, the paper designed an automatic image segmentation algorithm in order to stimulate its application in the testing segmentation precision. The experimental results for the low-noise image show that the correct rate of the proposed model is over 95% and the property is superior to the classical Fast linking model.
暂无评论