基于多特征的旋转不变纹理图像检索

为了消除图像旋转对图像检索的影响,提出了一个基于非下采样轮廓波变换、灰度共生矩阵和新相似性度量的旋转不变纹理图像检索算法。非下采样轮廓波变换具有各向异性和平移不变性。灰度共生矩阵描述了图像的方向性、邻近空间关系和方差的变化范围。通过计算每个非下采样轮廓波变换尺度下的所有子带的平均能量和平均标准差,灰度共生矩阵的二阶矩角、惯性熵、惯性矩、反差分矩,惯性相关系数的均值、标准差,得到具有旋转不变性的纹理多特征。提出一种新的相似性度量以改进纹理图像检索性能。实验结果表明,对于1个含640幅图像的旋转纹理图像库,与基于双树复小波变换的方法相比,该文图像检索算法将旋转纹理图像检索准确率由73.28%提高至80.71%。

Rotation-invariant Texture Image Retrieval Based on Multi-feature

In order to eliminate the effect of image rotation on image retrieval,a novel rotation-invariant texture image retrieval algorithm is presented based on the nonsubsampled contourlet transform(NSCT),gray level concurrence matrix(GLCM)and novel similarity measurement.The NSCT has anisotropy and translation invariability.The GLCM reflects the direction,adjacency spacing relationship and range of variance change of the image.The rotation-invariant features are achieved by calculating the average energy and average standard deviation of all subbands at each NSCT scale,the mean and covariance of the second moment angle,inertia entropy,inertia moment,contrast points moment of the GLCM.A novel similarity measure is presented to improve the retrieval performance of texture images.Experimental results demonstrate that:compared with the dual tree-complex wavelet transform based approach,the image retrieval algorithm improves the retrieval accuracy from 73.28% to 80.71% for the rotated database of 640 images.