摘要： 针对基于内容的图像检索所面临的图像低级视觉特征和高级语义之间的语义鸿沟问题,提出一种基于区域的图像检索算法。在LUV 颜色空间中使用K 均值聚类算法进行图像分割,提取分割后各区域的颜色、形状和区域自相关特征构成区域的综合特征,采用二次型距离相似性度量方法完成图像之间相似性的计算。实验结果表明,该算法具有较好的图像检索性能,与MIRROR 中各算法相比,使用平均归一化修正检索等级得到的检索性能提高了12% ~47. 8% 。

关键词: 基于区域的图像检索, 语义鸿沟, 区域自相关特征, 图像分割, 二次型距离相似性度量, 平均归一化修正检 索等级

Abstract: This paper presents a novel Region-based Image Retrieval (RBIR) method to reduce the semantic gap between low-level visual feature and high-level semantic of images in the content-based Image retrieval area. K-means clustering algorithm is used in the LUV color space for image segmentation. The color and shape feature in each region as well as the region auto-correlation feature are extracted as the integrating features of each region. And this paper defines a new Quadratic Distance Similarity Measure(QDSM) to calculate the similarity between different images. Experimental results show that the novel RBIR method using integrating features increases the retrieval performance by 12% ~47. 8% compared with the traditional methods under Average Normalized Modified Retrieval Rank(ANMRR) metric.

Key words: Region-based Image Retrieval(RBIR), semantic gap, region auto-correlation feature, image segmentation, Quadratic Distance Similarity Measure(QDSM), Average Normalized Modified Retrieval Rank(ANMRR)

中图分类号: 

• TP391. 41