People search by photographs
Every few days I get a message from users all over the world, with more or less the same question: I have a photo of someone I am looking for, can you help me find this person on the Internet? (This is usually referred to “people search” with photograph images, or “reverse image search”, or “similar image search”).
Unfortunately, I can’t. I’d need to have access to all images of all people of this world to use face recognition on them. Some Intelligence Service might have such a database, I don’t.
You need to have access to a real huge collection of images, and you need very strong computing power. That can only be done by companies with huge resources. Therefore, for some time it was only Google with such a capability. Their image search is not using face recognition (afaik), but their scale invariant feature technology works nice on face images, too.
Now that it is no longer Google alone to offer such a service, here is a list of image search engines that can compare images within a reasonably large database (send feedback if you know more):
- Google Image Search
Just upload a photo to Google’s Image Search, and it will find similar images. With Google owning probably the largest database of images on the web, chances are high that you find that (or at least a similar) photo.
- Bing Image Search
Bing now also offers an Image Search (using photo upload or URL), but it is harder to find than Google’s feature. Just use the link given above, then click the “Image Match” icon, and then “Upload an Image”.
- Yandex Reverse Image Search
Russian search engine Yandex also offers reverse image matching, including image upload.
- Baidu Reverse Image Search
Chinese search engine Baidu also offers reverse image matching with image upload.
- Karma Decay Image Search on Reddit
Beta version of an image search engine on Reddit.com.
- TinEye Reverse Image Search
Claims to have indexed over 12 billion images. Allows image input either via upload or with a given URL.
- PicTriev – Searching Faces on the Web
This one uses real face recognition to find a similar looking face. Unfortunately it searches only within a limited set of “celebrities”.
Some institutions allow their work to be tested by submitting an image to their face detector / face finder online demos:
- Twins Or Not? from Microsoft.
You can upload two photos and the face recognition engine will calculate a similarity score (100% means identical). This should be very useful for all of you who want to know whether a given face image represents the same person as in a different image. The face recognition engine is actually quite good – it is using Microsoft’s Project Oxford web service.
- How old do I look? from Microsoft.
You can upload a photo with any amount of faces on it, and the site will answer with all found faces marked, together with an age estimation of each person.
Don’t be too shocked if your estimated age is far off reality (+- 15 years accuracy is my rough guess), but the face detector seems to be highly accurate. Microsoft uses its Project Oxford web service for doing that, so you might want to check that out.
- Real-time Face Detector 2.0 from Alexander Telnykh. Detection of faces in frontal angle on digital photos and in videostreams, detection of faces in a tilt perspective. Allows front objects between ±15° from horizontal, detection of faces in profile perspective, demo of Eyes Localization, demo of Face Tracking, demo of skin-color filtering.
- Facefinding via Silverlight control from StarTrinity using Haar features / Adaboost.
Comes with sample code.
- Submit face images to the Betaface Live Demo by Oleksandr Kazakov (there is also a discussion group available)
- Submit face images to the University Illinois online face finder