Marcos J Víctor, Cristóbal Gabriel
J Opt Soc Am A Opt Image Sci Vis. 2013 Aug 1;30(8):1580-91. doi: 10.1364/JOSAA.30.001580.
In this paper, a method to characterize texture images based on discrete Tchebichef moments is presented. A global signature vector is derived from the moment matrix by taking into account both the magnitudes of the moments and their order. The performance of our method in several texture classification problems was compared with that achieved through other standard approaches. These include Haralick's gray-level co-occurrence matrices, Gabor filters, and local binary patterns. An extensive texture classification study was carried out by selecting images with different contents from the Brodatz, Outex, and VisTex databases. The results show that the proposed method is able to capture the essential information about texture, showing comparable or even higher performance than conventional procedures. Thus, it can be considered as an effective and competitive technique for texture characterization.
本文提出了一种基于离散切比雪夫矩来表征纹理图像的方法。通过考虑矩的大小及其阶数，从矩矩阵中导出一个全局特征向量。我们将该方法在几个纹理分类问题中的性能与通过其他标准方法所取得的性能进行了比较。这些方法包括哈勒克的灰度共生矩阵、伽柏滤波器和局部二值模式。通过从布罗达茨、Outex和VisTex数据库中选择具有不同内容的图像，进行了广泛的纹理分类研究。结果表明，所提出的方法能够捕捉有关纹理的基本信息，表现出与传统方法相当甚至更高的性能。因此，它可被视为一种用于纹理表征的有效且有竞争力的技术。
