一种基于改进的2D-Gabor纹理图像分类方法

针对多纹理图像分类的问题,本文提出了一种操作性强,通用性高的分类方法。借助人类视觉特性和纹理图像的尺寸,设计了一种快速简单的Gabor滤波参数设置方法。在多通道的滤波特征图像中应用顺序向前搜索策略选择特征,以J-M距离(Jeffreys-Matusitas distance)为判别因子进行特征空间的优化,最后通过SVM方法实现图像分类。实验表明,该方法有良好的纹理图像分类效果。较之传统的Gabor滤波图像分类方法,该方法具有参数设置简单,操作性强的特点。

A novel texture image classification algorithm via improved 2-D Gabor

In order to resolve the classification problem of multi-texture images, this paper proposed a novel classification algorithm with high operability and great generality. Based on the characteristics of human vision and the size of texture image, a simple and fast method was presented for the parameter setting of Gabor filters. The method began with feature selection of multi -channel filtered feature images based on sequential forward searching strategy. Then it optimized the feature space according to the Jeffreys-Matusitas distance. Finally it classified the images by SVM. Experimental results showed that the proposed method achieved good classification performance for texture images. Comparing with to traditional image classification method based on Gabor filters, this method was operable and simple in setting parameters.