Leveraging Non-Relevant Images To Enhance Image Retrieval Performance

Inherent subjectivity in user's perception of an image has motivated the use of relevance feedback (RF) in the image retrieval process.

RF techniques interactively determine the user's desired output or {\em query concept} based on user's relevance judgment on a set of images. Either a parametric or a non-parametric model for the user's query concept is typically employed. The parameters of the assumed model are estimated and refined based on user's relevance feedback. Parametric models offer higher robustness in estimation of parameters when the size of user's feedback is small. Non-parametric models permit easy incorporation of non-relevant images. Most parametric model based RF algorithms use only relevant images to refine the parameters of a distance metric. Consequently images in the database close to the non-relevant images continue to be retrieved in further iterations.

In this paper we propose a robust technique that utilizes non-relevant images to efficiently discover the relevant search region. A similarity metric, estimated using the relevant images is then used to rank and retrieve database images in the relevant region. A decision
Surface is determined to split the feature space into relevant and non-relevant regions. The decision surface is composed of hyperplanes, each of which is normal to the minimum distance vector from a non-relevant point to the convex hull of the relevant points. Experimental results demonstrate significant improvement in retrieval performance for the small feedback size
Scenario over two well established RF algorithms.