摘要： 针对采用单一方法提取图像特征时检索率不高的问题,结合非下采样剪切波变换(NSST)统计特征和旋转不变的局部相位量化(RI-LPQ)原理,提出一种纹理图像检索方法。非下采样剪切波不仅具有方向选择性及平移不变性,而且可以对图像进行有效的稀疏表示,与传统小波相比,可有效捕捉图像的边缘轮廓等纹理信息,与非下采样轮廓波相比,具有更高的计算效率。利用广义高斯分布函数对图像NSST 高频子带系数的统计特征进行分 析,RI-LPQ 描述算子直接提取图像特征,采用具有权重系数的相似性测度公式对Brodatz 图像库进行纹理图像检索。实验结果表明,与传统小波和轮廓波的方法相比,NSST 统计特征方法的平均检索率分别提高4． 77% 和1． 44% ,纹理图像检索方法的平均检索率分别提高7． 36% 和1． 98% 。

关键词: 非下采样剪切波变换, 广义高斯分布, 纹理图像检索, 旋转不变的局部相位量化, 特征提取, 特征融合

Abstract: For a single method of extracting image feature retrieval defective,this paper proposes a combination of Nonsubsampled Shearlet Transform (NSST) statistical features and Rotation Invariant Local Phase Quantization (RILPQ)texture image retrieval method. NSST exhibits highly directional sensitivity and shift invariance,even it can be sparse representation of the image. In contrast,NSST acquires the natural texture and edge information with the traditional wavelet and has higher computational efficiently with nonsubsampled contourlet transform. This paper acquires the statistical features of the image NSST coefficients by Generalized Gaussian Distribution(GGD) function. Image features are directly extracted by RI-LPQ description operator. Texture images on the Brodatz image database are retrieved by the formula of similarity measure with weight coefficients. Experimental results show that average retrieval rate of NSST statistical features method is 4． 77% and 1． 44% higher than traditional wavelet method and Contourlet method respectively. The average retrieval rate of this paper method based on fused features is 7． 36% and 1． 98% higher than NSST method and RI-LPQ method respectively.

Key words: Nonsubsampled Shearlet Transform (NSST), Generalized Gaussian Distribution (GGD), texture image retrieval, Rotation Invariant Local Phase Quantization(RI-LPQ), feature extraction, feature fusion

中图分类号: 

• TP391． 41