基于分形协方差函数的自然纹理描述与分类

Fractal-based covariance function description and classification of natural texture images

研究了自然纹理图像的描述与分类的方法,提出了基于分数布朗运动模型及其协方差函数的方法.分数布朗运动的协方差函数被用来估计自然纹理特征的Hurst系数和常数k .2个子图像的5个特征组成1 0个特征的特征集.与直接从原始纹理图像获得特征矢量不同,该方法的1 0个特征矢量是分别基于大于图像灰度平均值的图像和小于图像灰度平均值的图像得到的.以纹理图像的平均值为阈值,可以得到2幅子纹理图像.从每个子纹理图像提取出5个特征,它们分别是横向、纵向和45°方向的常数,横向和纵向距离为2的Hurst系数.2个子纹理的5个特征组成1 0个特征的特征集.从Brodatz纹理集选出的1 6种纹理图像被用来检验描述和分类效果,分类结果显示该方法具有很好的自然纹理的描述和分类能力.

To deal with the problem of charactering and classifying natural textures in images, a technique was employed which is based on the fractional Brownian motion model and its covariance function. The covariance function of fractional Brownian motion was proposed to estimate the Hurst coefficient and constant k which are used to character the natural textures. The feature set has ten feature vectors which are combined with both five features of two sub-texture image. Ten features were based on both the above average gray level image and the below average gray level image rather than based on the original image. Relying on the mean value of texture images, two sub-texture images were obtained. Five features were based on sub-texture image, the horizontal constant, the vertical constant, the 45° directional constant and the horizontal and vertical Hurst coefficient as the distance is 2.16 natural textures from the Brodatz album were considered, and the classification results show the efficiency of the technique.