A Rotation Invariant Descriptor Using Multi-directional and High-Order Gradients
In this paper,we propose a novel method to build a rotation invariant descriptor using multi-directional and high-order gradients(MDHOG).To this end,a new dense sampling strategy based on the local rotation invariant coordinate system is first introduced.This method gets more neighboring points of the sample point in the interest region so that the intensity distribution of the sample point neighborhood can be described better.Then,with this sampling strategy,we design the multi-directional strategy and use 1D Gaussian derivative filters to encode MDHOG for each sample point.The final descriptor is built using the histograms of MDHOG.We have carried out image matching and object recognition experiments based on some popular image databases.And the results demonstrate that the new descriptor has better performance than other commonly used local descriptors,such as SIFT,DAISY,MROGH,LIOP and so on.
Local descriptor Rotation invariant coordinate system Multi-directional strategy High-order gradients 1D Gaussian derivative SIFT
Hanlin Mo Qi Li You Hao He Zhang Hua Li
Key Lab of Intelligent Information Processing,Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China
国际会议
广州
英文
372-383
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）