特征符号随机场-Gibbs模型及其在纹理分割中的应用

FSRF-Gibbs Model for Texture Segmentation

本文提出了特征符号随机场的概念，定义了新的特征符号随机场-Gibbs模型，并讨论了它在纹理分割中的应用与传统马尔可夫随机场模型方法相比，由于包容了更多、更细致的图象信息，本文的方法能够得到更精确的分割结果，同时，新模型仍然有比较简单的模型形式，模型估计方法简单、利于在线运用.与传统特征聚类分割方法（如多通道特征聚类分割方法）相比，本文不要求得到相对纹理区域具有稳定性的特征，并利用Gibbs模型来描述空间变化的特征.从而使分割过程基于更本质的纹理特征，使分割结果更具普遍性.本文以标准Brodatz纹理为实验样本，取得了令人满意的实验结果.

This paper defines the Feature Symbol Random Field(FSRF), while presents a novel FSRF-Gibbs model for texture segmentation.The main function of FSRF is acting as 2D representation of texture feature vectors which come from the multichannel analysis. What should be emphasized is that all the employed features are spatial-changed, i. e. need not to be stable for certain texture region. By employment of FSRF, this poper also isgnificantly eases the problem in model estimation of Markov Random Field(MRF). As a result, finer and more reasonable segmentation is expected by involving both multichannel analysis techniques and fine MRF model. Finally, a new algorithm is included, which is easy to perform and leads to satisfactory experiment results on Bredatz Textures.