一种基于随机游走模型的融合视觉单词共现性的软分配词袋技术
Bag of Features with Co-occurrence Fusing Visual Words Mutual Information Based on Random Walk Model
张晋1
作者信息
- 1. 四川大学计算机学院，成都 610065
- 折叠
摘要
提取局部兴趣点并通过词袋技术表征图片，是进行图片检索的一种经典方案。在传统的词袋技术中，每一个局部兴趣点的分配都是相互独立的，而没有考虑相邻局部兴趣点分配的相互影响。这样可能会导致某些局部兴趣点不可靠的分配，从而降低图片检索的精确度。通过统计视觉单词的共现性，可以学习到一些有价值的先验知识；同时利用随机游走模型，将视觉单词的共现性融合到传统的软分配词袋技术中，从整体上减少局部兴趣点不可靠的分配，进而提升图片检索的精确度。
Abstract
A classic scheme to retrieve image is to extract local interesting points and use bag of features to represent an image. In traditional bag of features method, every local interesting point’s assignment is mutually independent, the scheme does not consider the interaction between two adjacent local interesting points’ assignment. So this can incur unreliable assignment for some local interesting points, and low the overall accuracy of image retrieve. Considers that learn some valuable priori knowledge by counting the co-occurring times between visual words; then fuses the co-occurrence information between visual words into the original soft-assigned bag of features based on random walk model, in order to avoid unreliable assignment of local interesting points and improve overall accuracy of image retrieve.
关键词
局部兴趣点/视觉单词/词袋技术/共现性/随机游走模型Key words
Local Interesting Point/Visual Word/Bag of Features/Mutual Information/Random Walk Model引用本文复制引用
出版年
2015
NETL
NSTL
维普
万方数据