Binary Descriptors for Dense Stereo Matching | ||||
International Journal of Intelligent Computing and Information Sciences | ||||
Article 9, Volume 21, Issue 2, July 2021, Page 124-139 PDF (1.92 MB) | ||||
Document Type: Original Article | ||||
DOI: 10.21608/ijicis.2021.63324.1073 | ||||
View on SCiNiTO | ||||
Authors | ||||
Hanaa Ibrahim Fariz Ibrahim 1; Heba Khaled 2; Noha Aly AbdElSabour Seada3; Hossam Faheem4 | ||||
1Computer Systems Department, Faculty of Computer and Information Sciences, Ain Shams University, Cairo, Egypt | ||||
2Department of Computer Systems, Faculty of Computer & Information Sciences, Ain Shams University, Abbasia, Cairo 11566, Egypt | ||||
3Lecturer at Faculty of Computer & Information Sciences, Computer Systems Department, Ain Shams University, Cairo , Egypt. | ||||
4Professor of Computer Systems, Computer Systems Department, Faculty of Computer and Information Sciences, Ain Shams University | ||||
Abstract | ||||
Dense local stereo matching is traditionally based on initial cost evaluation using a simple metric Dense local stereo matching is traditionally based on initial cost evaluation using a simple metric followed by sophisticated support aggregation. There is a high potential of replacing these simple metrics by robust binary descriptors. However, the available studies focus on comparing descriptors for sparse matching rather than the dense case of extracting a descriptor per each pixel. Therefore, this paper studies the design decisions of well-established binary descriptors such as BRIEF (Binary Robust Independent Elementary Features), ORB (Oriented FAST and rotated BRIEF), BRISK (Binary Robust Invariant Scalable Keypoints) and FREAK (Fast Retina Keypoint) to decide which one is more suitable for the dense matching case. The expremental results shows that agregation is required for use with binary descriptors to handle edges. Also, BRIEF produced the smoothnest disparity map if geometric transformations is not present. Whereas, FREAK and BRISK achieved the least overall error percentage across all regions. The lastest Middlebury Stereo benchmark is utilized in the experiments. | ||||
Keywords | ||||
Stereo Matching; Dense Matching; Binary Descriptor; Descriptor Matching; Computer Vision | ||||
Statistics Article View: 186 PDF Download: 260 |
||||