Kernel-Based Adaptive Image Retrieval Methods - Multimedia Database Retrieval: Technology and Applications - page 16

Chapter 2
Kernel-Based Adaptive Image Retrieval
Methods

Abstract This chapter presents machine learning methods for adaptive image
retrieval. In a retrieval session, a nonlinear kernel is applied to measure image
relevancy. Various new learning procedures are covered and applied specifically
for adaptive image retrieval applications. These include the adaptive radial basis
function (RBF) network, short term learning with the gradient-decent method,
and the fuzzy RBF network. These methods constitute the likelihood estimation
corresponding to visual content in a short-term relevance feedback (STRF). The
STRF component can be further incorporated in a fusion module with contextual
information in long-term relevance feedback (LTRF) using the Bayesian framework.
This substantially increases retrieval accuracy.

2.1 Introduction

Adaptation of the traditional similarity function plays a vital role in enhancing the
capability of image retrieval and broadening the domain of applications for machine
learning. In particular, it is often necessary to adapt the traditional Euclidean inner-
product to the more flexible and nonlinear inner products characterized by relevance
feedback parameters. The new inner products lead to a new similarity metric. As a
result, the image retrieval has to be necessarily conducted in a new space that is
adaptively re-defined in accordance with different user preferences. This implies a
greater flexibility for image retrieval. The topics addressed in this chapter are as
follows:
Section 2.2 will look into the linear kernel that is implemented through the query
adaptation method, metric adaptation method, and a combination of these methods.
In a linear-based adaptive retrieval system, the similarity score of a pair of vectors
may be represented by their inner product or Mahalanobis inner product.
Depending on the data cluster structure, either linear or nonlinear inner products
may be used to characterize the similarity metric between two vectors. The linear
metric would be adequate if the data distribution is relatively simple. To handle more
complex data distributions, it is often necessary to adopt nonlinear inner products
prescribed by nonlinear kernel functions, e.g., the Gaussian radial basis function
(RBF). Section 2.3 introduce a single-class RBF method for adaptive retrieval.