Using Self-Organizing Maps to Explore and Query Multimedia Collections (Bachelor Thesis, Finished)
With the ever-growing amount of devices that are capable of recording videos, the task of handling multimedia collections by hand has continuously become harder. Since this also means that such collections are rapidly growing in size, the use of multimedia retrieval systems, like vitrivr, that are capable of finding textual and visual similar items became increasingly important. Though vitrivr provides great support in finding items, it has rather limited support in exploration, so a new way to discover the content of a collection shall be implemented during this Bachelor’s Thesis. We approach this challenge by taking advantage of self-organizing maps that will help us sort and group items of a given collection together. Moreover, an opportunity is given to the end user to mark retrieved results as either positive or negative what will then be taken into account during further iterations. Consequently, the introduction of relevance feedback allows to set a focus during such exploration and query tasks improving the overall experience. When the sole implementation of the map functionality is utilized, the allotment of a right spot has turned out to be rather difficult as an easy way to find some related area is missing during initialization. However, evaluation results suggest that it can be seen as a powerful platform when it is used in combination with the existing toolset.
Start / End Dates
2020/03/09 - 2020/07/08