On June 18th, the latest research results of NM-Net: Mining reliable neighbors for robust feature correspondences was published in IEEE International Conference on Computer Vision and Pattern Recognition (CVPR) 2019. Zhao Chen, the first author and master student in 2017th grade, gave an oral report at the conference.
This paper proposed an image feature consistency screening algorithm based on compatible neighborhood filtering, which overcame the problem of poor compatibility in the geometric spatial neighborhood in the existing screening algorithm of image feature consistency. This study extracted consistent neighborhood information for potential correct matching in images through a compatibility-based neighborhood filtering algorithm, and further fused the extracted local neighborhood information by using convolutional neural networks and finally output all initial matches(whether correct or not). Compared with the existing screening algorithm, this method effectively extracts the local information with consistency around the matching line and makes full use of the deep learning-based convolutional neural network framework, finally obtaining sound filtering effects in databases of single consistency and multiple consistency.
CVPR is one of the world’s top conferences in the field of computer vision and pattern recognition. In the Google Academic Impact Rankings released in 2018, CVPR ranked 20th in all disciplines (journals and conferences participate in the sequencing), the highest among all subjects in electronics, computer science, and information science. The papers accepted by the conference represent the world’s most advanced level in the field of computer vision and pattern recognition. Zhao Chen’s thesis was completed under the guidance of his tutor Professor Cao Zhiguo and Professor Li Xin of West Virginia University. In the stage of blind review, the paper received unanimous acceptance from the three reviewers, and was finally selected as the oral report in the conference (oral, the recruitment rate is only 5%) by the organizing committee. It is the first oral article in our college adopted by CVPR as the first author is a master student.
![](/__local/A/FB/90/7AB83971678D8DF1E33E76709E1_3C896CBE_166BA.jpg)
![](/__local/8/14/1F/3698A4ACD8EC5956D2B972F04FF_C767A623_20621.png)