基于非下采样轮廓波的图像检索

Image retrieval based on nonsubsampled contourlet

提出一种基于非下采样轮廓波的图像检索方法.首先对图像进行预处理;然后利用非下采样轮廓波的平移不变特性,将像素点分为强边缘、弱边缘和噪声点三类;最后忽略噪声点,提取强边缘和弱边缘的各个子带的均值和方差等作为特征进行检索.实验结果显示,与轮廓波等方法相比,该方法具有更高的鲁棒性、检索效率更高且特征维数低.

This paper proposes a strategy for image retrieval based on nonsubsampled contourlet. First- ly, preprocessing the image, then, the pixels are divided into strong edges, weak edges and noises. Finally, the means and variables of nonsubsampled contourlet transform coeiIlcients matrix in different subbands and various directions of strong edges and weak edges were extracted to form the feature vec- tors for image retrieval. Compared to the contourlet transform, this approach can not only get higher accuracy, but also reduce the dimension of feature vectors, and is more robust.