Comments on: Face recognition: next step http://blog.landspurg.net/facae-recognition-next-step/ Technology, wireless, games...and more..... Wed, 19 Nov 2008 12:04:25 +0000 http://wordpress.org/?v=2.5 By: Eric Janszen http://blog.landspurg.net/facae-recognition-next-step/#comment-87 Eric Janszen Thu, 09 Feb 2006 23:42:35 +0000 http://blog.landspurg.net/?p=56#comment-87 Spent a lot of time looking at face recognition a while back. All claim near 100% accuracy with hardly any false positives. BS. And that’s just telling one face from another based on relative location of eyes, nose, mouth, etc. at a fixed distance, consitent lighting, etc. Else can’t tell a popsicle from a screwdriver, or a train a mile away from a tootsie roll a foot away. Check out the progress in machine vision over the past 20 years.... even slower than continuous voice recognition, still firmly stuck in 1995 (all still use various flavors of acoustic approximation + statistical analysis of phonemes based on dialect-specific language models). I can see auto-tagging working on the wireless GPS Camera of the Future... SearchTagDataItem21243523567; User: Fred Flintstone; Date: 2/8/06; Time: 12:23:09 AM EST; Location: Wall Street/Broadway Avenue, NYC; Description: Fat+man+ass+crack; User Comment: “Assholes! That’s my 3 year old son Freddy eating a cheeseburger!” Prediction: The best we can hope for from this kind of AutoTagging is 2,435,875,345 matches on image search of the term “asshole”. Anyone think this is serious technology? There’s a sucker born every minute. Spent a lot of time looking at face recognition a while back. All claim near 100% accuracy with hardly any false positives. BS. And that’s just telling one face from another based on relative location of eyes, nose, mouth, etc. at a fixed distance, consitent lighting, etc. Else can’t tell a popsicle from a screwdriver, or a train a mile away from a tootsie roll a foot away. Check out the progress in machine vision over the past 20 years…. even slower than continuous voice recognition, still firmly stuck in 1995 (all still use various flavors of acoustic approximation + statistical analysis of phonemes based on dialect-specific language models).

I can see auto-tagging working on the wireless GPS Camera of the Future…

SearchTagDataItem21243523567; User: Fred Flintstone; Date: 2/8/06; Time: 12:23:09 AM EST; Location: Wall Street/Broadway Avenue, NYC; Description: Fat+man+ass+crack; User Comment: “Assholes! That’s my 3 year old son Freddy eating a cheeseburger!”

Prediction: The best we can hope for from this kind of AutoTagging is 2,435,875,345 matches on image search of the term “asshole”.

Anyone think this is serious technology? There’s a sucker born every minute.

]]>