以能量和峰度为特征的轮廓波纹理检索算法

Sub-band Energy and Kurtosis Featured Contourlet Texture Retrieval Algorithm

轮廓波纹理图像检索系统检索率较低的一个重要原因在于纹理特征的提取.针对该问题,从数学角度分析了纹理图像在轮廓波变换域上的统计特性,证明了采用子带系数的能量和峰度作为特征能够更好地刻画纹理结构.在此理论基础上,构造了一种纹理图像检索算法.该算法采用子带系数的能量和峰度级联来构造特征向量,并采用Canberra距离作为相似度度量函数.实验结果表明:在特征向量长度、检索时间、所需存储空间相同的情况下,该检索算法比基于同样架构的已有轮廓波检索算法有更高的检索率.

One important reason of low retrieval rate of contourlet texture image retrieval system dues to the texture feature extraction. Focus on this problem, the statistical character of texture image in contourlet domain sub-bands co- efficients were analyzed from mathematical view, energy and kurtosis were proved to be the most suitable features for charaetering textures. According to this theory, a texture image retrieval algorithm was proposed. The algorithm cas- cades the energy and kurtosis of each sub-band in contourlet domain to form feature vectors and Canberra distance for similarity measurement. Experiments results showed the proposed image retrieval algorithm is superior to that of the original contourlet system with same length of feature vectors, retrieval time and memory.