一种新颖的多尺度无监督图像分割算法

A NOVEL AND MULTI-SCALE UNSUPERVISED ALGORITHM FOR IMAGE SEGMENTATION

吉伯斯随机场(Gibbs Random Fields,GRF)作为一种引入图像空间信息的先验模型已广泛运用于贝叶斯图像分割中。然而迄今为止,所涉及的这类先验模型往往仅体现为单一尺度上的马尔科夫性,而在多尺度意义上却未曾涉及。首次通过扩展传统单尺度意义上GRF模型到多尺度上,即多尺度吉伯斯随机场,从而圆满地解决这些难题。实验表明:所提出的模型算法有很好的鲁棒性,且易于实现对图像无监督的精确分割。

Gibbs Random Fields（GRF） is a popular prior model widely used in Bayesian segmentation due to its excellent property in describing the spatial information of image. But until now ,the classical approaches only describe the Markovian property of single-scale instead that of multi-scale. In this paper,a novel and unsupervised algorithm named multi-scale GRF that addresses these problems perfectly is proposed by extending the classical single-scale model of GRF to a multi-scale one at the first time. Experiments have shown that our algorithm presented in the paper has excellent robustness and esay to be used in unsupervised and precise segmentation.