基于基元自相关图的图像检索

Image Retrieval Based on Texton Autocorrelograms

提出一种新颖的颜色特征描述子,它能够在基元的基础上表达颜色信息的空间相关性。首先将彩色图像量化为64种颜色,并且对图像进行基元分析,最后利用统计信息来描述图像特征。实验结果表明,基元自相关图的性能优越于颜色直方图和颜色自相关图;基元自相关图的计算复杂度和特征维数均明显地小于颜色自相关图。

A new color descriptor is proposed in this paper. The descriptor can represent the spatial correlation of color based on textons. Firstly,quantify the original images into 64 colors and analyze textons context,and then calculate the statistical information of textons to describe image features. Image retrieval experimental results show that the performances of our proposed method are better than that of color histogram and color autocorrelograms,and the computational complexity and feature dimension are smaller.