Relevance feedback in Surfimage

Abstract

Relevance feedback is one of the strong components of Surfimage, the INRIA content-based image retrieval system. Relevance feedback is about learning from user interaction, and is useful in tasks like query refinement and multiple queries. We present two relevance feedback techniques currently implemented in Surf image.

Manifold learning using robust Graph Laplacian for interactive image search

“…In the first category, different density estimation methods are used in RF including non parametric Parzen windows [17], Gaussian mixture models [8], logistic regression [6] and novelty detectors [7,23]. In [8,9], the authors introduced a notion of relative judgment of the user, i.e., the response is not binary but a relative number measuring the relevance of a displayed set of images.…”

“…Different schemes exist in the literature for the purpose of RF [20,32] which are either based on density estimation [17,13] or discriminative training [28], depending respectively on the fact that they model the distribution and the topology of the positive and (possibly) the negative labeled images or they build a decision function which classifies the unlabeled data. In the first category, different density estimation methods are used in RF including non parametric Parzen windows [17], Gaussian mixture models [8], logistic regression [6] and novelty detectors [7,23].…”