K均值聚类分割的多特征图像检索方法

Multi-feature image retrieval method based on K-means clustering segmentation

从图像数据库中快速、准确地检索出所需要的图像,具有广泛的应用前景。针对使用单一图像特征难以准确表达图像之间的差异问题,提出了一种利用颜色聚类分割和形状特征提取的图像检索算法。选择符合人眼视觉特征的HSV空间,分别重组最能描述图像颜色特征的H分量和形状特征的V分量;用K均值聚类算法对两个分量进行聚类分割,得到目标物体;提取目标物体的Hu不变矩和傅里叶描述子来描述形状特征;用欧式距离进行相似度测量并用于图像检索中。采用不同类型图像进行实验,结果表明该算法优于使用单一特征和一般分割方法的图像检索技术。

It has a wide range of applications to retrieve the required image from the image database quickly and accurately. Considering the inaccuracy of single image features in expressing the difference between images, a new approach for image retrieval using color clustering segmentation and shape feature extraction is proposed in this paper. The perceptually uniform HSV space is employed. The H component and V component are restructured and clustered by K-means clustering algorithm. Target object is obtained. Concerning the shape information, Hu invariant moments and Fourier descriptors of the target object are extracted. Then Euclidean distance is adopted for similarity measure. Different types of images are used to experiment. The Experimental results show that the new algorithm is more effective than single features method and traditional segmentation method in image retrieval.