基于分块颜色核特征的图像检索方法

Image retrieval method based on block color kernel feature

颜色特征是图像最直观的特征,传统的颜色直方图虽然能够描述图像中各颜色分量的比例,然而却破坏了图像各颜色分量的空间位置分布信息,从而使得图像检索结果不准确。因此,提出一种分块颜色核特征的方法,使用重叠分块策略,强调图像中心的视觉内容,同时,设计颜色核函数将分块颜色直方图映射到核特征空间,更准确地描述图像颜色特征。经过实验证明,该方法与全局颜色直方图相比,分块颜色特征有较高的查准率和查全率。

Color feature is the most intuitive features of images. Even though the traditional color histogram can describe re?spective color component in image,spatial location distribution information of respective color component is destroyed. It makes the image retrieval results inaccurate. Because of this reason,a method of block color kernel feature is proposed,which uses overlap block strategy to emphasize visual content of image center zone. Block color histogram is mapped to kernel feature space by the designed color kernel function,so the image color features are described more accurately. The experimental results show that in comparison with global color histogram method and block color feature method,the proposed method has higher precision ratio and recall ratio.