基于多通道Gabor滤波器的纹理图像非监督分类

Unsupervised Texture Segmentation Via Multi-channel Gabor Filters

提出了一种快速、实用的Gabor滤波器设计方法.首先根据人类视觉特性选定Gabor滤波器的频率带宽参数,利用图像大小确定滤波器的中心频率,构造一组多通道的Gabor滤波器在时域和频域中提取图像多方向和多分辨率的Gabor能量特征,对特征空间进行非线性变换和Gauss滤波处理,最后通过K-means方法实现纹理图像的分割.实验结果表明,该方法能有较好的纹理分割效果.

In this paper,a fast and practical implementation of Gabor filter is proposed.First,it gests the bandwidth parameter and center frequency of Gabor filter based on human vision and the size of texture image,constructing the multichannel Gabor filters and extracting the Gabor multitemporal and multidirectional energy features of image in the time and frequency domain and implementing nonlinear transform and Gauss low-pass filtering in the feature space.By adopting k-means clustering algorithm,the partition of the image is realized.The results of experiment demonstrate that this approach has its advantage in image segmentation