Change Proposal for ISSUE-66

I have completed my Change Proposal for ISSUE-66 (Remove Image Heuristics paragraph from img Element Section):


In the absence of evidence (or implementation in browsers), the current draft suggests that image heuristics algorithms may be used to recover from images with missing @alt. The technology not only does not exist: it cannot exist.

It's possible that Google Goggles was the proposed implementation of image analysis heuristics here, but while it may perform well with well-defined objects with large databases of established imagery, it is not a cure-all for the problem of missing @alt -- and if it were, in fact, that good, it would be better positioned as a tool for the authoring process than in the browser.

The reason for this is that images themselves are only place markers for what the author intends to express. It is the author, then, and not the image, that is most responsible for determining what fits best as alternate content. And for this reason, no automated tool can possibly claim to sufficiently repair missing @alt content. This sentence only serves to make that less clear.

User agents may use any technology they choose to improve the user experience for users with disabilities. Such an implementation may in fact have positive effects, in certain cases. However, it is not necessary to specify this, particularly if by doing so the necessity of human-created @alt is made less than perfectly clear.

A second problem is that @alt, when contained within an anchor, is according to best practice used to describe the purpose of the link. Consider the following code, which is extremely common:

<img src="diskette.gif" alt="Save">

The intent of the graphic is to convey, through a common icon, that the control using this image saves current work. An image heuristic would not find "Save" but instead an image of 3.5" diskette. Replacing the image with a description is not necessarily a good indication of the author's intent, even when the description is accurate.

Finally, image analysis systems such as Google Goggles require enormous (as in upper terabytes to lower petabytes) databases of content in order to function at anything better than parlor-trick levels of efficacy. At the moment, the only available provider for such a service is Google. This is, in effect, a binding in the specification to a closed-source, single-provider service, which sets a dangerous precedent.

I have updated the ChangeProposals page with this one:


Received on Friday, 15 January 2010 22:21:27 UTC