基于纹理的旋转不变图像检索算法的研究

Rotation-invariant image retrieval algorithm based on texture

提出一种基于Gabor变换的旋转不变多尺度广义粗糙度特征向量并结合自适应加权距离进行纹理图像检索的方法。利用图像Gabor分解的幅度谱,依据多尺度空间局部能量分布、Hurst分形指数、方向差别来计算纹理特征向量,最后采用自适应加权的街区距离作为相似性准则。仿真结果表明,该算法对旋转纹理图像取得了很好的检索结果。

A rotation-invariant and multi-scale generalized roughness feature is presented and it was combinded with an adaptive weighted distance to apply in texture retrieval. Texture image was decomposed based on Gabor transformation, the rotation-invariant vector was formed by making use of the amplitude spectrum and combining with energy distributions, Hurst parameters and direction differences. Finally, the adaptive weighted distance was used to measure the similarity. Simulation results show that the proposed approach performed well in the rotation texture images.