基于统计模型和KL距离的纹理图像检索

为了进一步提高纹理图像的检索性能,提出了一种基于统计模型离的纹理特征提取算法。根据小波分解的特点,从小波系数角度出发,以每个子带的小波系数系数直方图分布特性作为纹理特征,采用混合高斯模型和一般高斯模型分别对低频和高频信息进行描述,利用最大似然估计规则将特征提取和相似计算结合起来,采用KL距离进行度量。与一般高斯模型方法比较,该算法具有检索精度高等特点。理论分析和在纹理图像检索的对比实验数据说明该算法在纹理特征提取方面的性能较一般高斯模型方法提高了5%。

Texture Image Retrieval Based on Statistical Model and KL Distance

In order to enhance the performance of the texture image retrieval, a new method based on the statistical model was presented, which had obtained by the statistical characteristics of the wavelet transforms (PDWT) from the wavelet coefficient histogram distribution as the texture feature using the characteristics of the wavelet decomposition and combining feature extraction with similarity measurement by ML rule using KL distance for image retrieval. The texture feature employ Gaussian mixture model(GMM) to described the low-frequency information and generalized Gaussian density model(GGD)to described the high-frequency information, this method is better than these of the pyramid discrete wavelet decomposition transforms (PDWT)under the same feature extraction method and the same similarity measure. In the contrast experiment result for image retrieval, the retrieval efficiency of our method is better than that of GGD and this method improved retrieval rate about 5%, compared with generalized Guassian density model.