Slashdot Mirror
Racist Facial Recognition Software
An anonymous reader writes "A black man found that his HP facial-tracking recognition software wouldn't work. Then he discovered it worked fine for a white co-worker. From the article: 'HP's Tony Welch thanked Desi and Wanda, the video's creators, and promised that he and the team at HP were looking into why the camera was behaving the way it was. "The technology we use is built on standard algorithms that measure the difference in intensity of contrast between the eyes and the upper cheek and nose," he said. "We believe that the camera might have difficulty 'seeing' contrast in conditions where there is insufficient foreground lighting."'"
Luckily, a similar situation was addressed and resolved in a Better off Ted episode.
I have worked with Viola-Jones before - it is extremely robust to lighting conditions. However what it is not robust to is angular changes (twisting of face sideways). It is possible that HP are using some kind of naiive algorithm to achieve face tracking, an easy one for e.g. is simple edge analysis for eye recognition (easier if you have a infrared emitter - to exploit the red eye effect) and then using this extrapolate facial dimensions then to facilitate tracking. It is possible they use Viola-Jones for the initial stage to locate the face region and then begin tracking, so even though the VJ tracker is very good, the processing further in the chain isn't so robust. Part of the reason for doing this might be that the VJ tracker is expensive in terms of compute cycles.