4 ms·

Ask HN: How to do face recognition

2 points by petervandijck 16y ago

If I'm building an iphoto-like app online, how can I get my server to do face recognition (recognize where a face is in a photo, and also recognize who (of your limited set of friends) this face might be)? It seems pretty simple if even cheap cameras do the first part, and most photo software does the second part quite well. Is there an open source or commercial package we can get? An API? Are the algorythms easy enough to implement something like this in a month or so?

petervandijck 16y ago
http://developers.face.com/docs/ http://developers.face.com/docs/ has anyone used that? Any good?
- edens 16y ago
  try it, it has a sandbox where you can try detect / train / recognize API calls. (full disclosure, I'm a co-founder @ face)
st3fan 16y ago
I have some history with Polar Rose. They are currently licensing their Face Detection and Matching engine. Either as a web service or as a library that you can include and use directly on your own servers. http://info.polarrose.com/solutions.php http://info.polarrose.com/solutions.php
Jun8 16y ago
You seem to be new to the topic since you are confusing face detection and face recognition. Face detection is a relatively simple affair, given an image determine a bounding box for each candidate face. OpenCV (http://opencv.willowgarage.com/wiki/ http://opencv.willowgarage.com/wiki/) is an excellent library with a lot of computer vision related functions. Out of the box it has an example program for face detection which performs quite well with the default Haar model files it comes with. Detection rates of higher than 95% are not uncommon for photos with clean faces. So, you can set up a fairly good face detector in a day or so. Face recognition, on the other hand, is a totally different beast. The bounding boxes you detect in the detection step are fed into classifiers that you should train. If you are new to this field I don't recommend you do this training or develop your own algorithms, since it's a complicated problem. You say most photo software does this, but generally these are offerings from large companies, e.g. Picasa, that have years of face recognition expertise behind that product. My recommendation is (i) to abandon the recognition idea (gulp! this may be hard to stomach but the energy may be better spent somewhere else, as this is very hard to get right) or (ii) if you insist, use an online API (e.g. http://developers.face.com/docs/ http://developers.face.com/docs/).

2 points by petervandijck 16y ago

petervandijck 16y ago

http://developers.face.com/docs/ http://developers.face.com/docs/ has anyone used that? Any good?

edens 16y ago
try it, it has a sandbox where you can try detect / train / recognize API calls. (full disclosure, I'm a co-founder @ face)

st3fan 16y ago

I have some history with Polar Rose. They are currently licensing their Face Detection and Matching engine. Either as a web service or as a library that you can include and use directly on your own servers. http://info.polarrose.com/solutions.php http://info.polarrose.com/solutions.php

Jun8 16y ago

You seem to be new to the topic since you are confusing face detection and face recognition. Face detection is a relatively simple affair, given an image determine a bounding box for each candidate face. OpenCV (http://opencv.willowgarage.com/wiki/ http://opencv.willowgarage.com/wiki/) is an excellent library with a lot of computer vision related functions. Out of the box it has an example program for face detection which performs quite well with the default Haar model files it comes with. Detection rates of higher than 95% are not uncommon for photos with clean faces. So, you can set up a fairly good face detector in a day or so. Face recognition, on the other hand, is a totally different beast. The bounding boxes you detect in the detection step are fed into classifiers that you should train. If you are new to this field I don't recommend you do this training or develop your own algorithms, since it's a complicated problem. You say most photo software does this, but generally these are offerings from large companies, e.g. Picasa, that have years of face recognition expertise behind that product. My recommendation is (i) to abandon the recognition idea (gulp! this may be hard to stomach but the energy may be better spent somewhere else, as this is very hard to get right) or (ii) if you insist, use an online API (e.g. http://developers.face.com/docs/ http://developers.face.com/docs/).