Recognition from collections of local features
Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
An image matching system for object recognition in scenes of varying complexity was constructed in Matlab for evaluating the recognition quality of two types of image features: SURF (Speeded-Up Robust Features) and SIFT (Scale Invariant Feature Transform) using the affine Hough matching algorithm for finding matches between training and test images.
In the experimental part of the thesis, the matching was algorithm tested for varying number of image features extracted from the train and test images, namely 1000, 2000 and 3000 interest points. These experiments were carried out against 9 objects of different complexity, including difficulties such as repeating patterns on the image, down and upscaling of the object, cluttered scenes, silhouette features, partly occluded object and multiple known objects in the scene.
The work provides the directions for improvement of the given view-based recognition algorithm and suggests other possible ways to perform the object matching with higher quality.
Place, publisher, year, edition, pages
2012. , 88 p.
IdentifiersURN: urn:nbn:se:kth:diva-103227OAI: oai:DiVA.org:kth-103227DiVA: diva2:559086
Subject / course
Master of Science - Computer Simulation for Science and Engineering
UppsokPhysics, Chemistry, Mathematics
Lindeberg, Tony, Professor
Hanke, Michael, Docent