Correcting Faulty Segmentation Labels

Simple item page
dc.contributor.advisorGrootjen, F.A.
dc.contributor.advisorKwisthout, J.H.P.
dc.contributor.authorBaarda, Koen
dc.date.issued2020-08-01
dc.description.abstractIn this study, a convolutional neural network is developed to correct errors in a database containing some label errors. The correction of errors is tested by comparing the output of the network to the output of a previous algorithm in order to investigate the use of neural networks for error correction problems. The database used was a medical database in which the labels were segmentations of pills wrapped in bags that were generated by the previous algorithm. The results suggested a convolutional network containing ve residual layers was unable to successfully correct errors generated by the previous algorithm because the network made errors that the original algorithm did not make. These results however are doubtful since the network did show signs of error correction. In further research that concerns how neural networks handle error correction in image segmentations, focus on other network structures such as U-net is advised.en_US
dc.embargo.lift10000-01-01
dc.embargo.typePermanent embargoen_US
dc.identifier.urihttps://theses.ubn.ru.nl/handle/123456789/12755
dc.language.isoenen_US
dc.thesis.facultyFaculteit der Sociale Wetenschappenen_US
dc.thesis.specialisationBachelor Artificial Intelligenceen_US
dc.thesis.studyprogrammeArtificial Intelligenceen_US
dc.thesis.typeBacheloren_US
dc.titleCorrecting Faulty Segmentation Labelsen_US
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
Baarda, K 4613147-2021.pdf
Size:
12.25 MB
Format:
Adobe Portable Document Format
Download
Collections
Faculteit der Sociale Wetenschappen