基于Gabor小波的频域快速纹理特征提取

摘要

针对常用的Gabor小波特征提取方法卷积运算量大、耗时长这一不足,采用了将空间卷积用频域乘积代替,然后直接在频域提取Gabor小波滤波后幅值图像的子带能量作为原图像特征的方法,节省了大量的计算时间,加快了特征提取速度.实验将Gabor小波变换提取的特征应用于纹理图像分类,结果表明,在保证识别率的情况下,在此采用的方法比常用的方法能明显提高识别速度,具有更好的实时性.%For the shortcoming of high computing-intension and time-consuming of the convolution in the popular strategy of Gabor wavelet based feature extraction, an approach which can reduce the computing time and speed up the feature extraction is presented. With the space convolution replaced by frequency-domain product, the sub-band energy of Gabor magnitude taken as texture features of the original image is directly calculated in the frequency domain. In the experiments of the latter* the features calculated by Gabor wavelet transform were used in texture classification. The results show that, in the case of ensuring the recognition rate, the approach can significantly improve the recognition speed and has better real-time operation.