【摘要】 提出了一种新的基于内容的图像检索方法.其关键技术是:(i)提出一种检索图像内容纹理统计特征的新方法.定义若干规范灰度像素模式基元;计算这些基元沿不同方向和不同跨距成对出现的概率,可以组成灰度-基元共生矩阵;该矩阵可用来描述图像纹理方面的特征.(ii)通过构建分类矩阵进行二叉树判别,扩展了SVM的多类分类功能.实验测试表明该法可行,且具有较好检索性能.更多还原

【Abstract】 A new approach to the contentbased image retrieval is presented. The key points of this approach are: (1) Some regular local gray structure patterns called gray primitives are predefined and a new image feature called gray primitive cooccurrence matrix is proposed, which is based on cumulative calculation of the cooccurrence of the same primitives in specified directions and scales across the tested image, and this matrix can characterize the gray structure pattern statistic feature; (2) The classification capability of SVM is extended from twoclass classifier to multipleclass classifier by selforganized sequential classification processing. The experiments show that the new approach using SVM as classifier and combining primitive pattern cooccurrence feature with other color and shape features as classifying feature vectors possesses good retrieval classification performance.更多还原