Curvelet感兴趣区域相关图的纹理图像检索

提出了一种基于感兴趣区域纹理图像的检索方法.对纹理图像进行Curvelet多尺度分解,根据各层子带的能量分布提取感兴趣区域,并量化感兴趣区域Curvelet系数,计算颜色自相关图而构造图像的特征向量.在Brodatz纹理库中的实验结果表明,与原有基于Curvelet的纹理图像检索方法相比,所提出的方法对纹理图像检索的效果更佳.

Region of Interest Based Texture Image Retrieval Using Curvelet Correlogram

A novel algorithm based on region of interest was presented for texture image retrieval. The proposed method was based on a combination of Curvelet multi resolution image decomposition and colour correlation histogram. First, Curvelet transform was applied for texture image multi-scale decomposition. Then, the region of interest was extracted based on the distribution of the energy of different scales, and a quantization step was used before computing auto correlograms of the region of interest Curvelet coefficients. Finally, index vectors were constructed for similarity comparison. Experiments based on Brodatz benchmark database show that the algorithm is more effective compared to other retrieval methods based on Curvelet in texture image retrieval.Key words：