摘要
针对图像检索时的旋转问题提出一种基于Log Gabor小波的旋转不变纹理特征。依据Log Gabor分解子带的均值及方差分布并结合广义Hurst分形特征对纹理进行描述,并在各尺度上采用主方向排序的方法对方向子带特征进行调整得到旋转不变特征向量。分别采用旋转纹理库和普通纹理库进行仿真实验,结果表明算法在两种情况下都取得了较好的检索效果。此外,在相同特征类型和图像库的条件下将本算法与Gabor算法结果进行对比,实验表明本算法的检索性能明显优于Gabor算法的检索性能。
A rotation-invariant texture feature based on Log Gabor wavelet was presented for texture image retrieval. Texture image was decomposed based on Log Gabor wavelet. The feature was formed by making use of the means and standard deviations of the spectrums and combined with Hurst parameters. Then using the maximum orientation to rearrange the vectors, a rotation-invariant texture feature was got. In order to demonstrate the efficiency of the method for the general texture images, the experiments were carried on the rotation texture database and general Brodatz database respectively. Simulation results show that the proposed approach performed well in both of the experiments and the performance of this method is obviously better than the Gabor algorithm.
出处
《光学技术》
CAS
CSCD
北大核心
2006年第1期92-94,97,共4页
Optical Technique
