Chamfer matching for image transformation estimation in capsule endoscopic videos | |
| Author | Pobporn Danvirutai |
| Call Number | AIT RSPR no.CS-13-07 |
| Subject(s) | Capsule endoscopy Gastrointestinal system |
| Note | A research submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Computer Science, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Research studies project report ; no. CS-13-07 |
| Abstract | Wireless capsule endoscopy (W CE) is an alternative non-invasive method for imaging the digestive tract. A pill—sized capsule is swallowed and allowed to travel through the digestive system while it captures video frames. The resulting video may contain about 50000 or more frames, therefore taking hours for the GI doctor to analyze. A three-dimensional reconstruction of those video frames could help doctors determine the structure and position of abnormalities easily. However, in order to realize a reliable 3D reconstruction, problems such as camera movement (sometimes abrupt or backward) and non—rigidity of the scene need to be addressed. Moreover, the low frame rate (3—1 8 fps with 201 1 technology) can cause the camera to miss parts of the scene, resulting in incomplete 3D reconstructed views. Feature matching between frames is required to localize camera and determine depth in a 3D reconstruction. Chamfer matching is one method for contour matching between two images. It allows some degree of deformation (rotation, scaling, translation, and non-rigidity) between two images of the same contour to be match able. An initial set of experiments on pairs of correlated images shown that chamfer matching gives ?2% correct matches out of all candidate correspondences, after which refinement is required to select the best and most consistent correspondences. I add an outlier detection and match refinement phase using global chamfer distance mimization among candidate matches, where the penalties for matching with large translations and rotation, or short contours have been imposed. I compute camera translation, sealing, and rotation parameters for each match. Refinement of the transformation parameters into to obtain an optimal set of matches set of matches between two scenes was carried out for 49 image pairs, giving 77.5% true positive in matching contour of the selected scenes (38 out of 49 image pairs are correct). More than half of the matching errors are in frames in which the camera motion accelerates, making the images less correlated- For the more correlated scenes, the refined matching method gives promisingly correct results. |
| Year | 2013 |
| Corresponding Series Added Entry | Asian Institute of Technology. Research studies project report ; no. CS-13-07 |
| Type | Research Study Project Report (RSPR) |
| School | School of Engineering and Technology (SET) |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Computer Science (CS) |
| Chairperson(s) | Dailey, Matthew N.; |
| Examination Committee(s) | Mongkol Ekpanyapong;Guha, Sumanta; |
| Scholarship Donor(s) | Royal Thai Government Fellowship; |
| Degree | Research report (M. Sc.) - Asian Institute of Technology, 2013 |