一种基于多通道Gabor滤波的纹理提取方法

Texture extracting approach based on multichannel Gabor filters

研究了一种在多种纹理叠加的复杂轮胎X光图像中进行纹理提取的新方法。轮胎X光纹理在频谱中被看作是局部空间频率的辐射模型,并且不同纹理对应的主频率聚集区域基本能相互分开。基于Gabor滤波器对空间频率的带通滤波能力以及方向、形状、带宽和通带中心频率的可调节特性,构造了多通道Gabor滤波器组提取每种纹理对应的主要频率成分,提取的频率成分经过频-时域变换就实现了纹理提取。考虑到实际纹理灰度分布不是理想的正弦分布,改进了传统的Gabor滤波方式,使其保留了纹理的低频成分。实验结果表明,图像中叠加的多种纹理能被准确提取,纹理区域边界也没有失真,并且有效降低了纹理之间频谱混迭造成的影响。

A novel method was proposed for extracting each texture in complex X-ray multi-texture tire images. The textures were modeled as irradiance patterns containing a limited range of spatial frequencies, where mutual distinct textures differ from their dominant characterizing frequencies. Based on the properties of 2-D Gabor filter, such as the band-pass of spatial frequency, tunable orientation, tunable filter shape, tunable center frequency, and tunable bandwidth, muhiehannel Gabor filters were constructed to segregate each texture frequency domain. Then, each texture could be acquired by the channel amplitude responses. Considering the slight variation of gray level and non-ideal sinusoidal distribution in the inner region of texture primitives, traditional Gabor filters were improved to preserve the lower frequency components in texture spectrum. The experimental results showed that the textures had been truly extracted, and the boundary distortion had been removed. Also the influence of spectrum aliasing had been effectively reduced.