计算机与现代化

• 图像处理 • 上一篇    下一篇

 融合多种内容特征的图像检索算法

  

  1. 1.重庆大学计算机学院,重庆400044;2.软件理论与技术重庆市重点实验室,重庆400044
  • 收稿日期:2015-04-01 出版日期:2015-09-21 发布日期:2015-09-24
  • 作者简介:朱凌云(1992-),女,安徽安庆人,重庆大学计算机学院硕士研究生,研究方向:图像处理与检索; 朱征宇(1959-),男,安徽马鞍山人,重庆大学计算机学院、软件理论与技术重庆市重点实验室教授,博士, 研究方向:Web智能检索,图像处理与检索; 齐新勇(1988-),男,河南人,硕士研究生,研究方向:图像处理与检索。
  • 基金资助:
     国家科技支撑计划重点项目(2011BAH25B04)

 An Image Retrieval Algorithm Combining Multi-content Features

  1. 1. College of Computer Science, Chongqing University, Chongqing 400044, China;

     2. Chongqing Key Laboratory of Software Theory and Technology, Chongqing 400044, China
  • Received:2015-04-01 Online:2015-09-21 Published:2015-09-24

摘要:

为提高图像检索的准确度,提出一种基于内容的图像检索算法FCTL,它融合了图像的颜色、纹理和空间位置3种内容特征,生成一个300柄的特征直方图,用特征直方图的Matsushita距离衡量
图像间的相似度,按图像库中图像与搜索图像的相似程度排序返回检索结果。在WANG、UCID两大常用图像库上进行对比实验,并选用MPEG-7中推荐的ANMRR评价方法,实验结果表明,相比一些常用的图像
检索算法,FCTL的准确度有所提升。

关键词:  , 基于内容的图像检索, 直方图, 颜色, 纹理, 空间位置

Abstract:

In order to enhance the accuracy of image retrieval, a new content_based image retrieval algorithm FCTL is presented in this paper. In FCTL, the feature of an
image is described by a 300 bin histogram, which contains three types of information: color, texture and location. To represent the similarity of two images, a Matsushita
distance between their histograms is used. The retrieve results will be returned, sorting by their similarities with a given querying image. ANMRR, which is recommended by MPEG
-7, is used to evaluate performance in our comparative experiments. The comparative results based on WANG and UCID image libraries show that, FCTL has higher retrieval accuracy
than some frequently used algorithms.

Key words: contend-based image retrieval (CBIR), histogram, color, texture, location