logo logo logo

Thesis project Proposals


The table below presents some suggested master thesis projects based on the UiB project goals.

Note that aspects from different proposals may be combined into a single project. Note too that other projects are also possible within the CAIM project.

CAIM Goal UiB project areas UiB focus Thesis topic proposals


Dynamic context capture and management

1.a Image context model Definition of the concept of image context,

Context modeling including development of context descriptors for relationships among image objects

  1. Which context characteristics are most effective for image retrieval?
  2. Develop a metadata model suitable for capturing image context data in various application domains.
  3. How effective are ontologies when used for image classification, information retrieval and collection updates more effective?
  4. What role can folksonomies have in image description and retrieval?
  5. Can image context descriptors be utilized for semantic image description?

1.b Image description management Dynamic update of context descriptors from user feedback.

Development of techniques for the construction and use of ‘folksonomies’ and ontologies.

  1. How can user input (feedback) be utilized for object identification and inter-object relationship specifications?
  2. How do user defined 'folksonomies' relate to/effect application/domain ontologies?
  3. How will constant update of image descriptors, for example through use of social tagging, improve image retrieval?
  4. Can a game-based approach to identifying objects within an image improve identification of image components?


Multimodal information retrieval

2.a Context-based Image retrieval Focus on internal image contexts, e.g. relationships between image objects
  1. Develop/extend algorithms for image retrieval using context data as defined in 1a and 1b
2.b Image retrieval using visual queries Development of a visual query language (VQL) for image retrieval
  1. How can current VQLs be improved?
2.c Multi-modal information retrieval Text/audio retrieval from image input
  1. Given visual input, how can image context data be utilized to retrieve related, multiple modal (text, audio) information?
2.d Multi-DB information retrieval Focus on retrieval from multiple image collections.
  1. How can current information retrieval algorithms that can address multiple DB systems be adapted for a multiple media DB environment?
    (related to 2c)


Result presentation

3.a Context-based ranking Focus on image component context
  1. Develop and evaluate algorithms for "context-based" ranking
    (Related to 1b)
3.b Result ranking for image retrieval Focus on ranking/integration of image result sets from multiple image DBs
  1. Develop/improve algorithms for merging/ranking result sets from multiple image DBs (re. 2d)
3.c Multi-modal result presentation Mixed media presentation.

  1. Develop/improve current multi-modal result presentations Potential case studies:
  2. Are these systems domain dependent?
  3. How can these systems be adapted for mobile phone units?


Application development and evaluation

4 Application evaluation Identification of the feature set needed for effective image retrieval from mobile devices

A number of image retrieval prototypes exist within the caim project.

  1. Which prototype features are most effective for image retrieval?
  2. Which context descriptors are most effective for image retrieval?
  3. Which features and functions give the best support for users?

Modified: 08.01.2008 © University of Bergen - by: [an error occurred while processing this directive] Lars-Jacob Hove