Experimental evidence accumulated over the past years indicated the importance of the ranking process in information retrieval. As for the textual documents, image ranking is a task that involves different parameters. They depend on the intrinsic characteristics of an image, but also on the indexing language used for representing its semantic content. We developed a weighting model that combines these parameters in a general scheme. Finding the best balance between the parameters is not straight-forward. Different parameter combinations leads to different rankings, which may be more or less accepted by the users. In this paper, we choose a set of test queries and present the impact of the parameters on the rank of each image. Different combinations are discussed, and the best combination is specified. For the evaluation, we follow a user-oriented approach, and compare the ranking provided by each parameter combination to the ranking given by human judgment. This is a step toward a user-centered image retrieval system, which will dynamically adapt to the user's profile and preferences.
[1]
Philip J. Barnard,et al.
Modelling multimodal interaction: A theory-based technique for design analysis support
,
1997,
INTERACT.
[2]
Mourad Mechkour,et al.
EMIR2: An Extended Model for Image Representation and Retrieval
,
1995,
DEXA.
[3]
Iadh Ounis.
A Flexible Weighting Scheme for Multimedia Documents
,
1999,
DEXA.
[4]
Iadh Ounis,et al.
Modeling, Indexing and Retrieving Images Using Conceptual Graphs
,
1998,
DEXA.
[5]
Seiki Inoue,et al.
Evaluating the location of hot spots in interactive scenes using the 3R toolbox
,
1998,
CHI.
[6]
Iadh Ounis,et al.
RELIEF: combining expressiveness and rapidity into a single system
,
1998,
SIGIR '98.
[7]
J. J. Rocchio,et al.
Relevance feedback in information retrieval
,
1971
.