消除旋转影响是纹理图像特征提取及分类的关键问题之一.文中基于一类正交分段多项式函数系——U系统,提出一种纹理图像分类算法.首先通过U系统函数与三角函数的张量积定义了一类单位圆盘上的U调和基函数;基于该基函数,提出了图像的旋转不变U变换RIUTs,并在此基础上构造了具有平移、旋转及缩放不变的纹理图像描述子,从而得到纹理图像的特征向量.RIUTs既可以消除图像旋转的影响,又特别适合纹理图像特征提取.在通用纹理数据库中的纹理图像分类实验结果,表明了文中算法的优越性能.
How to deal with rotation effect in feature extraction and classification for texture images is one of the key problems. Based on a class of orthogonal piecewise polynomial function named as U system, this paper proposed a novel algorithm for texture classification. We first constructed a new basis functions on the unit disk by combining U system functions and trigonometric functions. Based on them, we defined the rotation-invariant U transforms(RIUTs) and then obtained the corresponding texture descriptors which are invariant for rotation, scale and translation transforms. Our method not only can eliminate the rotation effect, but also is very suitable for texture feature extraction. The results of the experiments show that our method have superior performance.