Abstract
Fast and accurate image segmentation is still a challenging task in computer vision because real-world images are often distorted by noise and intensity inhomogeneity. In order to overcome these problems, fuzzy clustering is extensively applied to image segmentation because of the strong ability to reject local minimum. It also incorporates with kernel metrics to enhance robustness against noise and outliers and construct a nonlinear energy function based on a variational level set framework. However, level set implementation often costs a lot of CPU times, and needs “re-initialization”or regularizing terms to keep level set function smooth and stable. Our research provides the algorithm which solve level set equation of kernel fuzzy active contour model by using Lattice Boltzmann Method solver. Lattice Boltzmann Method can recover the level set PDE by computing pixel-by-pixel independently and maintain the stability and smoothness of level set function in the meanwhile. Therefore, this algorithm is fast, stable and independent to the position of the initial curve. Experiments on synthetic and real-world images demonstrate the stability and performance of the proposed method, and also show the efficiency of algorithm by using graphics processing unit.