Contributions to local feature extraction, description and matching in 2d images
- BARANDIARAN MARTIRENA, IÑIGO
- Marcos Nieto Doncel Director/a
- Manuel Graña Romay Director
Universidad de defensa: Universidad del País Vasco - Euskal Herriko Unibertsitatea
Fecha de defensa: 12 de julio de 2013
- Richard J. Duro Fernández Presidente/a
- Ivan Macia Oliver Secretario/a
- Bogdan Raducanu Vocal
- Ramón Ferreiro García Vocal
- Javier de Lope Asiaín Vocal
Tipo: Tesis
Resumen
Nowadays,. Computer Vision is becoming a very important research topic because of its great applicability and usefulness in many and heterogeneous areas such as medical or bio-medical, astronomy, industrial,or educational sectors, as well as in entertainment industry or even in our every day life.Despite this variety of application areas, a great number of Computer Vision based applications integrates at some point of their processing pipeline. The identification, extraction and matching of some type local features across images. Local features are well suited to image recognition and matching because of robustness against noise and geometric or photometric transformations, providing concise representations of objects in the image. Several interest point detectors and local feature descriptors, as well as strategies and algorithms for matching them, have been presented since in the last decade. Though a lot of progress has been done in this field, the problem of matching points across different images is far to be fully solved..This thesis aims to contribute to the field of local image feature extraction and matching by giving useful insight of state-of-the-art, serving as a supplement to existing comparative studies about interest point extraction, feature description and matching, as well as by contributing with some new approaches regarding this technologies, such as a new local image descriptor based on the trace transform. We also contribute to the field by providing the scientific community with a verified and well designed tool and image data sets, that allow comparing results obtained from different approaches regarding interest point extraction, feature descriptor or descriptor matching.