TY - GEN
T1 - A dense correspondences matching algorithm with hierarchical pyramid architecture
AU - Sun, Zhe
AU - Zhang, Yong
AU - Chang, Qutong
N1 - Publisher Copyright:
© 2017 IEEE.
PY - 2017/7/2
Y1 - 2017/7/2
N2 - We introduce a novel matching algorithm for computing dense pixel correspondences, which is inspired from deep convolutional neural networks. Whereas the prevailing approaches operate at the pixel level, we propose a hierarchical, multi-layer pyramid architecture that compute the match at multiple scales-ranging from an entire image, to coarse grid cells, to every atomic patch. The algorithm consists of two steps: in the bottom-up mode, calculate the similarity of patches between two images in different scales; in the top-down mode, aggregate the similarity from the top layer until the bottom. We evaluate the performance of our algorithm on the Mikolajczyk and Middlebury datasets. Our model shows excellent results in particular for repetitive textures as well as the varying illumination.
AB - We introduce a novel matching algorithm for computing dense pixel correspondences, which is inspired from deep convolutional neural networks. Whereas the prevailing approaches operate at the pixel level, we propose a hierarchical, multi-layer pyramid architecture that compute the match at multiple scales-ranging from an entire image, to coarse grid cells, to every atomic patch. The algorithm consists of two steps: in the bottom-up mode, calculate the similarity of patches between two images in different scales; in the top-down mode, aggregate the similarity from the top layer until the bottom. We evaluate the performance of our algorithm on the Mikolajczyk and Middlebury datasets. Our model shows excellent results in particular for repetitive textures as well as the varying illumination.
KW - coarse-to-fine
KW - dense scene correspondence
KW - hierarchical pyramid architecture
UR - https://www.scopus.com/pages/publications/85047627102
U2 - 10.1109/CISP-BMEI.2017.8301991
DO - 10.1109/CISP-BMEI.2017.8301991
M3 - 会议稿件
AN - SCOPUS:85047627102
T3 - Proceedings - 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics, CISP-BMEI 2017
SP - 1
EP - 6
BT - Proceedings - 2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics, CISP-BMEI 2017
A2 - Li, Qingli
A2 - Wang, Lipo
A2 - Zhou, Mei
A2 - Sun, Li
A2 - Qiu, Song
A2 - Liu, Hongying
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics, CISP-BMEI 2017
Y2 - 14 October 2017 through 16 October 2017
ER -