中文摘要：

针对基于内容的图像检索系统图像纹理特征提取对图像检索精度存在影响的问题,提出一种基于曲波（Curvelet）变换与高斯混合模型（Gaussian Mixture Model,GMM）相结合的方法提取图像的纹理特征。该方法通过曲波对图像进行多尺度分析并结合K-means和期望最大化（Expectation Maximization,EM）算法来估计高斯混合模型的参数,以此构建图像的纹理特征空间。仿真结果表明,所提出的方法比传统的图像纹理特征提取方法精度更高,并且提高了图像检索系统中的检索精度。

英文摘要：

Since the image retrieval precision was affected by the image texture feature extraction of content-based image retrieval system,this paper proposes an improved approach which combines the Curvelet transform with the Gaussian Mixture Model to extract image texture feature.Combined with the algorithm of K-means and Expectation Maximization,it provides a way of analyzing the image in multiscale in the Curvelet Domain to construct the image texture feature space.The simulation results demonstrate that the improved method not only has higher accuracy than the traditional methods,but also can promote the retrieval precision.