一种基于空间特征的图像检索法研究

A Spatial Feature for Content-based Image Retrieval

在基于内容的图像识别技术中,颜色自相关图是一种常用的图像颜色特征提取法。文章在此基础上,考虑图像结构特征,通过对图像进行水平和垂直两次扫描,分别建立图像相似区域的变化图。该算法和颜色自相关图一并应用到SOM网络上,实现了一个基于颜色特征和结构特征的图像分类。编程实验结果表明:新算法可以较好地提高基于内容的图像检索的准确性,避免了仅靠图像的颜色特征进行分类和搜索的不能描述图像宏观结构的缺点。

Many approaches for extracting color feature are widely used for content-based image retrieval.Each of them has its unique advantages.However,we cannot present the visual characterization of an image comprehensively by simply using a color feature extracting method.In this paper,we present an algorithm to collect the spatial feature of an image. This algorithm creates two graphs through two line scans on the image,one of which is in horizontal direction and the other is in vertical.A system is implemented to test the efficiency of this new spatial feature extracting algorithm.By combining a popular color feature approach,called color auto-correlogram,we demonstrate that this new algorithm can associate with color feature well to enhance the accuracy of image indexing.