The Community for Technology Leaders
2016 International Conference on Frontiers of Information Technology (FIT) (2016)
Islamabad, Pakistan
Dec. 19, 2016 to Dec. 21, 2016
ISBN: 978-1-5090-5300-1
pp: 252-257
ABSTRACT
Scene Completion is an interesting Image Processing problem that has recently been studied in the context of data, i.e. by using large repositories of data. One of the requirements for such a data intensive approach is that the completion has to be done without human intervention. This is rather challenging as it may not be clear that what could be the most suitable image for the completion purpose in the data repository that potentially contains millions of images. We propose a methodology for finding the top-1 image in a data repository that could be the best candidate for scene completion. We do so by computing a representative set of features namely Gist, Texture and Colour, and then give an algorithm for scene completion. To obtain the top-1 image, we consider a ranking scheme that satisfies the value-invariance property and thus, is not affected by the individual feature scores. The scene completion algorithm completes the input image with the constraint that the completion has to be seamless. The approach is data-driven and there is no need of labelling by the user. Although, the completion process is automated, we also allow the user to select a completion image from the top-k matches in order to have a completion that is semantically valid. The experimental results show that we are able to find a matching image that is able to complete the input image seamlessly.
INDEX TERMS
Image segmentation, Image color analysis, Image retrieval, Feature extraction, Computer science, Electronic mail,Data Intensive Scene Completion, Scene Completion, Top-k Queries
CITATION
Romana Talat, Muhammad Muzammal, Imran Siddiqi, "Scene Completion Using Top-1 Similar Image", 2016 International Conference on Frontiers of Information Technology (FIT), vol. 00, no. , pp. 252-257, 2016, doi:10.1109/FIT.2016.053
94 ms
(Ver 3.3 (11022016))