Nonparametric Active Contour for Tubular Object Extraction
Yu Wang, Zulong Yu, Kaiqiong Sun · 2012
In this paper, we propose a nonparametric active contour model to extract tubular object with the application in segmentation of vascular angiogram. There is no tuning parameter between data term and prior term in the contour model and a full automated extraction method is obtained. The gray intensity of object and background are fitted by local image information, which contributes to the data term of the active contour model and makes the model can deal with intensity inhomogeneity. The prior term of the model consists of curve normalization which smoothes the curve as in normal region-based active contour model. The smoothing term, however, is not based on the curve length associated a tuning parameter which need to be geared as in the classic active contour model. Instead, a probability distribution of the contour point associated with the curve contributes to the prior energy, which depends on a local scale at each pixel that can be estimated. The results on synthetic image compared with other methods are presented.