Local statistic information-driven active contours for image segmentation
Xiaosheng Yu, Qi Qi, Nan Hu · 2016
This paper presents a local statistic information-driven active contour model for image segmentation. The local intensities of each inhomogeneous object are modeled as Gaussian distributions. The means of Gaussian distributions in each local region are modeled as a bias field multiplying the piecewise constant which reflects the physical property of inhomogeneous objects. A local statistic information fitting energy functional is presented with the level set function, the bias field, the piecewise constant, and variances as variables. The proposed model is implemented by an efficient numerical schema to ensure sufficient numerical accuracy. It is validated on numerous synthetic images and real images, and the promising experimental results show its advantages in terms of robustness and accuracy.