| Modeling Concept Learning
Process for Image Retrieval
Presented by: Anlei Dong
ABSTRACT : Relevance feedback reduces the gap between low-level
visual features and high-level human concepts in image retrieval.
However, it is only the adaptation of feedback received from
a single user in response to a query image. Exploiting previous
retrieval experiences from multiple users further helps to
learn visual concepts. Currently, this learning process is
only empirical without a mathematical model. This paper proposes
a statistical model consisting of overlapped clusters which
are modeled as Gaussian density functions. Based on this model,
we analyze how retrieval performance is improved with relevance
feedback and retrieval experience. Both simulation and real
data results support the theory.