基于多尺度及多方向分析的纹理图像检索算法

Texture Image Retrieval Algorithm Based on Analysis of Multi-scale and Multi-orientation

提出一种基于多尺度及多方向分析的纹理图像检索算法.针对纹理图像的特点,从不同尺度和方向比较了实数离散小波变换和双树复小波变换的性能.充分利用双树复小波变换的旋转不变性、良好的方向选择性以及有限的冗余等优点,将其有效地应用于纹理特征提取过程中.通过提取各子带上的能量和标准差作为特征矢量,采用Camberra距离作为相似度量进行检索,减小了计算量,取得了良好的检索效果.实验结果表明,双树复小波提取纹理特征所获得的检索性能优于实数离散小波检索算法,也优于经典的灰度共生矩阵算法,且算法具有良好的旋转不变性.

A novel image retrieval algorithm based on analysis of multi-scale and multi-orientation is proposed.Aimed at the characteristics of texture image,the performance of real discrete wavelet transform and dual tree complex wavelet transform is compared.The excellent characteristics of rotation-invariance,good orientation selection and finite redundancy are fully utilized,and applied in texture feature extraction.The algorithm uses energy and standard deviation on each subband of the decomposed image as feature...