Abstract: Edge is an effective feature for similar texture classification. For better improving the edge detection accuracy, Varied Local Edge Pattern（VLEP） algorithm is used to extract the texture edge feature in which multi-scale and multi-direction properties are taken into account. Generally, texture feature extraction is based on regional statistics of the intensity change between the pixel and its neighbors. However, large deviation often exists, especially when the image resolution changes, or when the image is lighted and reflected. Therefore, in this paper dominant learning framework approach based on the VLEP is proposed to classify similar texture images, which constructs global dominant set to solve the problem of small inter-class and large intra-class distance resulting from the above deviation. The experimental results show that the proposed method can effectively improve the classification performance of similar texture image.

Key words: texture classification, Varied Local Edge Pattern（VLEP）, dominant learning framework, global dominant pattern set

摘要： 边缘是进行相似纹理图像分类的有效特征之一，为了提高边缘检测精度，使用可变化的局部边缘模式（Varied Local Edge Pattern，VLEP）算法，利用像元及其近邻的灰度变化进行区域统计，同时从多尺度和多方向的角度提取纹理边缘特征。然而，当图像分辨率发生变化，或图像受到光照、反射的影响时，纹理计算可能会出现较大偏差。为此，在VLEP算法的基础上，提出主导学习框架相似纹理分类方法，通过构建全局主导模式集，解决纹理计算偏差导致的类间距离小和类内距离大的问题。实验结果表明，主导边缘模式思想可以有效地提高相似纹理图像的分类准确率。

关键词: 纹理分类, 可变局部边缘模式, 主导学习框架, 全局主导模式集