计算机与现代化 ›› 2015, Vol. 0 ›› Issue (4): 82-85,89.

• 图像处理 • 上一篇    下一篇

考虑视觉显著性的部分相似图像检索

  

  1. 南京理工大学计算机科学与工程学院,江苏南京210094
  • 收稿日期:2014-12-19 出版日期:2015-04-27 发布日期:2015-04-29
  • 作者简介: 陈秋平(1988-),男,江苏丹阳人,南京理工大学计算机科学与工程学院硕士研究生,研究方向:图像处理,图像检索。

 Partial-duplicate Image Retrieval via Exploiting Visual Saliency

  1. School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
  • Received:2014-12-19 Online:2015-04-27 Published:2015-04-29

摘要:

当前流行的图像检索系统普遍采用词袋(Bag-of-Words)模型表示图像视觉内容。由于传统的视觉词袋模型忽略了局部特征间的几何关系,考虑几何约束的后处理方法明显地提高了检索准
确率。这些方法认为每个局部特征点是平等的,然而在实际情况中,图像中的局部特征点对于部分相似图像检索任务的重要性是不同的,比如位于相似图像区域上的特征点要比位于背景图像区域的特征
点重要。鉴于此,提出考虑图像特征点重要性的部分相似图像检索算法。首先用视觉显著性算法来计算图像每个像素点的显著性(即重要性),然后在几何验证计算图像间匹配分数中考虑匹配局部特征
点的重要性,最后在广泛使用的相似图像检索数据集上对提出的算法进行验证。实验结果表明了本方法的优越性。

关键词:  , 图像检索, 相似图像, 视觉相似性, 几何验证

Abstract:

 Bag-of-Word (BoW) model is widely used to represent the visual content of images in state-of-the-art image retrieval systems. Since traditional BoW models
discard geometric relationships among local features, exploiting geometric constraints as post-processing steps has been demonstrated to significantly improve the performance.
Each local feature is treated equally in these methods. In practice, the importance of local features for image retrieval is different. For example, the local features in the
duplicate image regions are more important that ones in the background image regions. Towards this end, a new partial-duplicate image retrieval method is proposed by leveraging
the importance of local features. First, visual saliency method is introduced to calculate the importance of each pixel, and then the importance is used to weight the similarity
of matched points. To verify the performance of the proposed method, extensive experiments are conducted on the widely-used datasets and experimental results show the
outperformance of the propose method.

Key words: image retrieval, duplicate image, visual saliency, geometry verification