Improving Image Search

Images are notorious for being among the most challenging objects to search for on the Internet and in library catalogs and databases. Unlike printed works, they typically don’t identify their creator, publication date, or the people, places or events that they visually represent. Creating detailed bibliographic records for images requires painstaking research and expensive cataloging.

 

An image record from Harvard’s Visual Information Access Catalog

A new technology has recently been developed by engineers at UC San Diego that promises to aid the process of searching for images on the Web and in digital projects. Supervised Multiclass Labeling (SML), automatically analyses the content of images, compares it to various “learned” objects and classes, and then assigns searchable labels or keywords to the images. SML can also be used to identify content and generate keywords for different parts of the same image.

In addition to significantly improving image searching on the Web and supplementing the detailed cataloging of image collections by libraries and museums, automated image content analysis technologies like SML could potentially have a number of interesting research applications.

Imagine, for example, being able to perform an image content analysis of Marc Chagall’s paintings, examining the frequency of his religious and folk iconography during different periods of his life. If we think of an artist’s “language” as being expressed through images and symbols, then automated content analysis might some day be used by art historians and researchers as a kind of visual concordance for better understanding art objects, artists, and movements.

The next revolution would then be to apply similar content analysis tools to moving images and videos.

 

Resources:
Article & Video: New Algorithms from UCSD Improve Automated Image Labeling

Article Abstract: Supervised Learning of Semantic Classes for Image Annotation and Retrieval