On the relevance of colour features in multimedia retrieval (Master Project, Ongoing)

Author

Rahel Arnold

Description

Cineast, the retrieval engine of the content-based, multimedia retrieval system vitrivr supports colour features since its earliest iterations.

However, in recent instalments of interactive multimedia retrieval evaluation campaigns, such as the Video Browser Showdown (VBS) and Lifelog Search Challenge (LSC) such colour features rarely contributed to a successful task resolution.

Nevertheless, colour information is often used in descriptions or tasks of such evaluations. 

To still leverage this colour information, query input as well as feature extraction should be systematically analysed and the results of the analysis might prove useful in the context of staged queries.

Thus, in a primary step, the current colour feature extraction modules of Cineast have to be systematically investigated, also in the context of human colour perception.

In a second step, the investigation's results will be used to optimise query formulation as well as feature extraction.

In a subsequent step, new extraction modules should be designed and implemented, in order to enhance query-by-sketch.

An interesting approach for this might be a pre-processing step to emulate sketches (e.g. by super-pixel aggregation in combination with other measures).

Start / End Dates

2021/08/23 - 2021/12/30

Supervisors

Research Topics