School of Engineering

Graduates the engineering leaders of tomorrow...

Unsupervised Multi-label Region-based Image Annotation

The daily access to social media platforms such as Facebook, Instagram, Twitter and Pinterest, has generated a huge number of images with user defined captions. These large amount of labeled images have led to the execution of multiple studies that discuss different techniques to annotate these images. Our study suggests a region-based unsupervised image annotation tool that assigns textual descriptions from previously annotated images on social media. The offline training phase consists of creating visual feature clusters with their textual annotation by implementing hierarchical clustering methods combining high-level semantic features and lowlevel visual features. Then, new images can be annotated by applying similarity based classification between the features of the new image and the visual feature clusters built during training.  Report in pdf
19-20-SP-ImageAnnot-JT-res.jpg

Copyright 1997–2021 Lebanese American University, Lebanon.
Contact LAU | Feedback