Resource title

Image representations and feature selection for multimedia database search

Resource image

image for OpenScout resource :: Image representations and feature selection for multimedia database search

Resource description

The success of a multimedia information system depends heavily on the way the data is represented. Although there are “natural” ways to represent numerical data, it is not clear what is a good way to represent multimedia data, such as images, video, or sound. In this paper the authors investigate various image representations where the quality of the representation is judged based on how well a system for searching through an image database can perform. The system is based on a machine learning method to develop object detection models from example images that can subsequently be used for example to detect – search – images of a particular object in an image database. The authors design and experiment with image representations for leaning to detect objects, in particular faces and people in images – although the same techniques and representations can be used for other types of object detection tasks or multimedia data analysis problem. They first consider classes of linear transformations of the images. In particular, they present an experimental comparison for object detection using as image representations raw pixel values, projections onto principal components, and Haar wavelets. They also show experiments which indicate the dramatic benefit of histogram equalization for face and people detection, a non-linear transformation, and discuss a theoretical explanation for this result. As a base classifier for the detection task, they use Support Vector Machines (SVM), a kernel based learning method. Within the framework of kernel classifiers the authors investigate new image representations derived from probabilistic models of the class of images considered, through the choice of the kernel of the SVM. Finally, they present a new feature selection method using SVM. Multimedia data are typically represented with a large number of dimensions resulting to high processing times for any task. Their feature selection algorithm is used to choose only a small number of the dimensions with which to represent the images without significant loses in terms of the performance of the detection – search – system.

Resource author

Resource publisher

Resource publish date

Resource language


Resource content type


Resource resource URL

Resource license

Copyright INSEAD. All rights reserved