基于分数布朗运动和概率神经网络的自然纹理分类

该文基于自然纹理的分数布朗运动(FBM)模型,引入一种新的分形特征参数——分形尺度,提出计算分形尺度的截矩线性度(IAL)方法。实验表明,将纹理图像的分形尺度与分形维数结合,既反映了物理表面分形的尺度范围又反映了该尺度范围内的纹理粗糙程度,从而更有效地描述和区分纹理特征。文中利用纹理图像在水平、垂直和对角3个方向上的分形尺度和分形维数作为特征集,采用概率神经网络(PNN)作为分类器进行自然纹理图像分类,得到准确的分类结果。

Texture Classification Using Fractional Brownian Motion and Probabilistic Neural Network

In this paper, the fractal scale is introduced as a new measure to describe natural physical surfaces. An Intercept And Linearity (IAL) method is proposed to determine the fractal scale automatically. Experiments indicate that the fractal scale, together with the fractal dimension can effectively reflect the fractal scale range of natural textures and the texture roughness within the scale range. For the purpose of classification, the fractal scales and the fractal dimensions along horizontal, vertical and diagonal directions are obtained to form a feature set with six parameters. The probabilistic neural network is used as the texture classifier. Results have shown satisfied classification of natural textures.