基于PDTDFB域混合建模的纹理图像检索

Texture Image Retrieval Based on Hybrid Modeling in PDTDFB

为了联合考虑特征提取和相似性测度以提高图像检索系统的性能,提出一种基于金字塔双树方向滤波器组(PDTDFB)域混合建模的纹理图像检索方法.首先将高频子带系数建模为广义高斯分布,将复方向带通子带的幅值系数建模为Weibull分布,并通过改进的最大似然估计方法确定其分布参数;然后采用一种新的相似性测度分别对VisTex和Brodatz数据库的纹理图像进行检索,得到和查询图像相似的图像子集.实验结果表明,与采用单一统计建模的方法相比较,该方法有效地提高了纹理图像的平均检索率.

For improving the performance of image retrieval systems, feature extraction and similarity measurement are used jointly. In this paper, a method of texture image retrieval which is based on hybrid modeling in the pyramid dual-tree directional filter bank （PDTDFB） is proposed. First, coefficients of the high frequency subbands are modeled as generalized Gaussian distribution, and the magnitude coefficients of the complex directional bandpass subbands are modeled as Weibull distribution. The parameters are estimated by an improved maximum likelihood method. Then, the texture images of the VisTex and Brodatz databases are retrieved respectively based on a new similarity measurement. Accordingly the image subsets similar to the query images are selected. The experimental results demonstrate that the proposed method can improve the average retrieval rate of the texture images effectively, comparing with the approaches using the single statistical modeling.