基于双密度Contourlet变换的图像检索

提出了一种新的双密度Contourlet变换,理论证明该变换在L2(Z2)空间是框架算子,具有较低平移敏感性和多方向分辨能力的优点。纹理图像在该变换域的高频方向子带系数分布符合广义高斯分布,可以利用广义高斯参数估计表征图像高频子带的纹理特性;针对变换域的低频子带,采用局部二值模式(LBP)提取图像的局部纹理特征。基于内容的图像检索实验表明,所提算法检索精度比传统Contourlet变换算法提高了5.3%。

Image Retrieval Using the Double Density Contourlet Transform

We proposed the double density contourlet transform(DDCT) and studied its applications. The DDCT is a frame operator for L2(Z2)and has the advantages of near shift invariant, multi-scale and multi-direction. The DDCT sulrbands of the texture images have the property of non-Gaussian. The generalized Gaussian density(GGD) can denote the whole statistical feature of image to some extent At the same time, the local binary pattern is used to describe the local texturcspatial feature for the low frequency sulrband of multiwavelets. I}hus the GGD and local binary pattern features can be computed as the feature texture. Experiments indicate that the retrieval efficiency of this algorithm is raised by 5. 3% than the Contourlet algorithm.