Replies

The key is that you just have to do your crunching in the right place.

I've worked with pattern-recognition software, in particular that by Excalibur Technologies, as well as fingerprint recognition software, and it really does work. And that was getting on towards 5-7 years ago. They may not be using Excalibur's neural-net algorithms, but there are other approaches that work too.

Fingerprint recognition algorithms don't compare the prints bit-for-bit - they build algorithmic maps of the fingerprint image, summarizing the print in several different points corresponding to key features, like loops, arches, and whorls. The same approach is commonly used in facial recognition software. This serves to condense an image that may be thousands of bits in size down to a concise summary that can be quickly and efficiently compared for near matches and solid hits, and quickly and efficiently transmitted over limited-bandwidth links.

Since every one of the potential hits will be on the same day, there's no concern about aging, facial hair, etc, which simplifies the problem significantly.

Here's the steps, as I would construct them:

1. Your high-speed laptop, with a video-in port, monitors the incoming video stream and locates and maps faces of individuals entering the polling place by taking a snap every few seconds. The software to do this kind of video and/or image processing is established and while relatively cutting-edge, is available and well-understood.

2. You store a local copy of each face image, and its corresponding map summary, under a unique identifier tagged with the laptop's name and a sequential number. You transmit the map summary and the ID number over the wireless link to the central database.

3. The central database receives the map and ID data, and stores it along with all the other map and ID data from the other polling stations. Each incoming map is compared with all the other stored maps.

4. If the central computer finds a close match or a hit on a particular face-map summary, it sends a request back to the laptop that had the first submission of the map, based on its ID number, and uploads the actual video image of the face in question, then transmits that full image down to the laptop that just sent in the matching face-map summary.

5. A little bell goes "ding" on the laptop that sent in the most recent image, and the operator compares the two images - the one just taken on the local laptop, and the one just downloaded from the central computer.

6. If there's a match to the human eye, the operator clicks the "print" button, and a color printer spits out a page that is then handed to one of the assistants at the polling place, who then goes to make a citizens arrest of the culprit and calls the police.

This approach deals with the limited bandwitdh of the CDPD network by sending only the summarized, compressed face map data up, and not sending the full image data until actual vote fraud is suspected.

You seriously underestimate the ingenuity of the avearage software engineer, my friend.

PS - I'm not involved in the project, this is just the first gloss approach I'd take if I were.