基于分级显著信息的空间编码方法

传统的Bag of Words模型检索方法并不具备局部特征间的空间关系,因此影响检索性能.本文提出了基于分级显著信息的空间编码方法.通过分层次的提取显著区域并对每个显著区域内的特征点进行空间编码.目的是探索特征间的空间关系,并根据分级显著信息提高特征间的相关性.在几何验证过程中,本文通过任意三点间的角度编码和位移编码构成的空间编码方法完成图像对之间的空间关系匹配,同时根据图像各个区域间的显著程度赋予该区域空间关系匹配得分相应权重,得到最终的几何得分,重新排列检索结果.实验结果表明本文提出的方法既改善了最终检索结果的精确度又降低了几何验证阶段的计算时间.

Spatial Encoding Based on Hierarchical Salient Information

The traditional model "Bag of Words" does not capture the spatial relationship among local features,thus affecting the retrieval performance.Hence,the spatial encoding method based on hierarchical salient information is proposed,which aims at fully exploring the geometric context of all visual words in images and increasing the discriminative power of the features based on hierarchical salient information.We propose a new encoding method in the geometric verification step.The spatial layout of every 3 points within a certain salient area will be represented by angle encoding and location encoding,meanwhile we sum all spatial matching scores with weights based on hierarchical salient information to generate the final ranking list.Experimental results prove that our scheme improves the retrieval accuracy significantly and reduces the computing time during the geometric verification step.