Learning Transformations for Exemplar Based Matching
Event details
| Date | 17.06.2019 |
| Hour | 09:30 › 11:30 |
| Speaker | Vidit Vidit |
| Location | |
| Category | Conferences - Seminars |
EDIC candidacy exam
Exam president: Prof. Pierre Dillenbourg
Thesis advisor: Prof. Pascal Fua
Thesis co-advisor: Dr. Mathieu Salzman
Co-examiner: Prof. Ronan Boulic
Abstract
In computer vision, finding similarity between two images is useful for several tasks like, object detection and segmentation, tracking, image retrieval, image registration, etc. The task in exemplar based matching is to find regions in two images, similar to one presented by the exemplar. This relatively easy sounding task has been one of the challenges in the field.
The difficulty arises from the fact that exemplar can undergo several geometric and photo-metric transformations, which makes it visually quite different in appearance to the image where the match is to be made. Several approaches have tried to formulate such transformations upto a certain limit. This research proposal is aimed at mitigating challenges faced in previous works with the help of modern machine learning methods.
Background papers
Exam president: Prof. Pierre Dillenbourg
Thesis advisor: Prof. Pascal Fua
Thesis co-advisor: Dr. Mathieu Salzman
Co-examiner: Prof. Ronan Boulic
Abstract
In computer vision, finding similarity between two images is useful for several tasks like, object detection and segmentation, tracking, image retrieval, image registration, etc. The task in exemplar based matching is to find regions in two images, similar to one presented by the exemplar. This relatively easy sounding task has been one of the challenges in the field.
The difficulty arises from the fact that exemplar can undergo several geometric and photo-metric transformations, which makes it visually quite different in appearance to the image where the match is to be made. Several approaches have tried to formulate such transformations upto a certain limit. This research proposal is aimed at mitigating challenges faced in previous works with the help of modern machine learning methods.
Background papers
Practical information
- General public
- Free