Pattern Recognition

How would an AI parse sensory input from sound or sight? For example, how would one make a program to count the number of humans in a given picture?

 

Log in or Sign up to hide all adverts.

Here's a good overview site, with links to open source software and libraries: Object Recognition in Image Processing

Here are some searches on a few academic sites, for nasty cutting edge details. "Image processing" and "object recognition" seem to be the standard terms:

Google Scholar
Science Direct
Communications of the ACM

 

Log in or Sign up to hide all adverts.

this is a tough one. companies like microsoft,ibm, and google have been working on things like this for years and theeres don't work very well.

 

Log in or Sign up to hide all adverts.

This is actually one of the fields where the mental apparatus vastly outperforms computers at present.

 

If it is already known that a given picture contains humans you are in with a chance to count the blobs, but if the hope is for recognition as a human would understand it, that as yet is all but non existent except with a set up designed for the purpose, with a face or fingerprint especially placed for attention.

Accurate voice recognition and synthesis is also notoriously tricky. I know of somebody, a friend's father, who spent a lifetime on that but never yet finished the work.

 

even a simple thing like still pictures is hard to make out what images are in it. identifiying a human image from a picture is only something humans can do. a computer needs to know what a human looks like first before it can delineate the human from the picture.

currently, the pattern recognitions are expert systems where they predefine what are objects that it should look for. for face recognition its the face, for licenses plates its licenses plates, for fingerprints its fingerprints.