Motivation

In modern computer displays, window managers are ignorant of the visual structure of a user's display. While the manager knows where each window is, it knows nothing about the actual contents of the window. Moreover, without knowing the specific task the user is performing, the manager does not know which areas of the screen are relevant. This means that pop-ups and other alerts often appear over information which the user is attempting to utilize. In addition to being distracting and annoying, this concealment of relevant information makes the user's task more difficult and frustrating.

This difficulty of determining which elements of a display are important information also applies very directly to the Internet. Bad web page design can obscure vital content with poor contrast. This can make the page difficult to use for normal users, and almost impossible for users with impaired vision (eg, sufferers of macular degeneration). Current approaches to this problem use a "knife switch" methodology: they strip the page of all color, converting it to a black and white view. While this certainly improves contrast, it also completely destroys the designer's visual design

Approach

We are exploring the problem of content identification on web pages. By exploiting the structure of the Document Object Model (DOM), we can use component level techniques to isolate textual information within a page. The background and foreground colors can then be adjusted according to a user-specified contrast preference. Thus, we transition from a "knife switch" (all or nothing) contrast adjustment model to an "analog dial" (gradient) model for contrast enhancement. This system will provide a test bed for models and algorithms for content separation.

Moving from the Internet context to the realm of the desktop, we will use the techniques and models developed for the web framework to locate regions of importance in the user's display. From this information, we will attempt to construct a "relevancy mask" which can be used both to improve image contrast and to intelligently guide pop-up placement.