摘要： 为提高纹理图像的检索效率，提出一种基于统计模型的纹理特征提取方法。采用金字塔对偶树方向滤波器组实现图像变换，利用Gamma分布对方向子带系数进行建模，通过矩估计方法得到分布参数，并作为纹理检索的特征，使用改进的KL距离度量相似性。在 VisTex彩色纹理图像数据库上的检索结果表明，与DT-CWT小波变换相比，该方法的平均检索率较高。

关键词: 金字塔对偶树方向滤波器组, KL距离, 纹理特征, 图像检索, Gamma分布

Abstract: To enhance the performance of texture image retrieval, a new method based on statistical model is presented. It proposes a color texture retrieval method using Pyramidal Dual-tree Directional Filter Bank(PDTDFB) decomposition to gain multi-scales and multi-directions sub-bands information. For extracting feather from sub-bands coefficients and measuring similarity, the Gamma distribution and KL distance are provided. Experimental results from VisTex texture images database show that the average retrieval rate of proposed method is higher than DT-CWT method.

Key words: Pyramidal Dual-tree Directional Filter Bank(PDTDFB), KL distance, texture feature, image retrieval, Gamma distribution

中图分类号: 

• TN911.73