基于Gabor滤波器的织物纹理图像分类

摘要

In view of strong directivity and specific periodicity shown in the fabric texture images, an approach is proposed to extract the feature of fabric texture images with Gabor filters.Firstly, the Gabor filters are used to analyze the fabric texture images from different directions and scales.Then the mean and variance of the filtered results are calculated to characterize the textural features.Finally, an one-vs-all scheme is applied to train multiple Support Vector Machine (SVM) classifiers for the classification of different types of fabric texture images.Experimental results show that the proposed method performs well both in representing the fabric textural features and providing promising classification performance.%针对织物纹理具有很强的方向性和周期性特点,基于Gabor滤波器织物纹理图像特征,提出了一种实现纹理分类的方法.该方法首先利用Gabor滤波器从不同尺度、不同方向对织物纹理图像进行分析,然后计算滤波结果的均值和方差表征纹理特征,最后使用一对多方案训练多个支持向量机(Support Vector Machine, SVM)分类器对不同类型织物纹理图像进行分类.实验结果表明,提出的方法能够有效表征织物纹理图像特征,具有很好的分类性能.