一种基于颜色特征的图像检索方法  

A Color Based Image Retrieval Method

作　　者：黄元元[1] 何云峰[2] 

机构地区：[1]南京航空航天大学计算机科学与技术系,南京210016 [2]南京理工大学精密仪器系,南京210094

出　　处：《中国图象图形学报》2006年第12期1768-1773,T0003,共7页Journal of Image and Graphics

摘　　要：提出了一种新的基于颜色自组织聚类的图像检索算法。该算法根据颜色在图像中的分布信息,通过对像素点的加权处理确定最佳的颜色初始聚类数目,将经过加权处理的像素点进行聚类,得到图像主要颜色的统计特征,同时,在分析了主要颜色在图像中的散布情况后,利用主色的矩特征来描述各个主色在图像中的空间分布特征,结合颜色的统计特征与空间分布特征来描述图像的内容,保证了图像在局部和整体上的一致性,实验结果证明,该算法的检索精度明显要高于传统的仅基于颜色统计特征的检索精度。By means of self-organizing clustering, a new color-based image retrieval method is proposed in the paper. According to the colors＇ distributing information in the image, every pixel is assigned a weighing value and thus the initial number of clustering can be confirmed. Therefore, those weighed pixels are clustered and the dominant colors＇ statistical features are acquired. Based on the dominant colors spread in the image, the colors＇ moment features are extracted to present their special features simultaneously. Therefore, the whole image＇ s content can be expressed from general statistic to partial distributing by these two kinds of features. The experiments verify the method mentioned above is more efficiently than those ways based on color histogram.

关 键 词：基于内容的图像检索 颜色特征 聚类 矩特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]