一种局部聚合描述符和组显著编码相结合的编码方法

A new feature coding algorithm based on the combination of group salient coding and VLAD

局部聚合描述符(vector of locally aggregated descriptors,VLAD)的特征编码方法在大规模图像检索上取得了较好的效果。但是,VLAD存在硬分配难以准确描述局部特征向量与视觉词汇隶属关系的问题,本文将两种软分配编码与VLAD相结合来增强局部特征向量与视觉词汇的隶属关系。新的编码方法在15 Scenes、Corel 10和UIIC Sports Event数据库上的实验结果表明:1)在VLAD中加入局部软分配能够提高分类准确率,而且对比Fisher编码在分类准确率上也有一定的优越性;2)除了软分配,显著性对提高分类准确率也起到了一定的作用。

The vector of locally aggregated descriptors （VLAD） has achieved good results in addressing large-scale image retrieval problems; however, VLAD has a defect in that the relationship between local descriptors and visual words cannot be accurately described using hard assignments. In this paper, we therefore combine two kinds of soft assignment coding methods with VLAD to enhance the relationship between local feature vectors and visual words. We applied our method to 15 scenes from the Corel 10 and UIUC Sports Event datasets, with our experimental re- suits showing that our combined partial soft assignment coding method and VLAD was able to enhance classification accuracy and achieve better classification accuracy than the well-known Fisher Coding approach. In addition to soft assignment, saliency also plays an important role in enhancing classification accuracy.