"Ontology Based Tagging Project" Web Page

About

The aim of the project is to create an application which will be used as a test bench. It must combine different methods of image analysis and tags suggestion and provide a way to obtain and save the results so they can be compared later. The main goal is the use of ontologies to avoid the semantic gap produced with the low level comparison methods.

The application combines the following:

Image low level feature extraction: Extracting feature vectors representing the content of an image, so it can be compared with the others of a database to search for similarities. These vectors are about color (Color Coherence Vector, Color Histograms, ...), shape (Eccentricity, Geometric Moment, ...) or textures (Primitive Length Vector).
Comparison of feature vectors: Second phase is to compare an image vector with the others of the database to search for the similar ones. To do this we use Euclidean Distance between two vectors or the LSI (Latent Semantic Indexing) algorythm so we have a value in the range [-1.0, 1.0], using 1.0 when two images are equal.
Getting the tags: Last phase is tagging the image. To do this we use the comparison value obtained at the previous step. We propose different methods to suggest tags by weighting them: some using the similarity values between images and others using an ontology. The aim of the project is to use the ontology so it accurates the tags by filtering them and suggesting others related with the initial concepts. The initial concepts are weighted using the similarity values between images.
Users: Other aspect still in development stage are the users tags. An image can have different tags for different users, as well as their importance in that image can differ depending on the user.

This project was created to be used as the thesis of the Computing Engineer University Degree at the 'Universidad Politécnica de Ingeniería de Gijón' of the Oviedo University, Spain.

The title of the thesis is 'Automatic Extraction of Semantics from Images Based in Collaborative Tagging and Ontologies' and was created to be part of the PhD thesis of the lecturer Doina Ana Cernea, Mathematics Department, Oviedo University. The PhD thesis title is 'Integración de las etiquetas colaborativas de los usuarios en sistemas automáticos de indexado semántico de imágenes'. Tutors of the thesis are José Emilio Labra Gayo and Doina Ana Cernea.