In the most recent update to its iPhoto application, Apple offered up a relatively noninformative description of the changes it had made: better "overall stability" and fixes for "minor issues in a number of areas, including Faces, Places, photo sharing, and slideshows."
What Apple didn't spell out in its release notes is that the update includes a slew of enhancements -- some of them quite useful, if not immediately obvious -- to the latest version of iPhoto, which was unveiled in January at the Macworld Expo. In particular, Apple cranked out improvements for the app's cool facial recognition and location/geocoding features known, respectively, as Faces and Places. I'm already a fan of iLife '09 and have used the iPhoto features for the past few weeks.
But many iPhoto users may not yet have tried them out or even be aware of the recent changes. So here are the details.
Detect/Rescan for missing faces
One of the problems with Faces in its initial release is that it occasionally wouldn't recognize a person's face as a face. This was particularly common in photos where someone's face is turned to one side or was shot in low light, or in slightly blurry photos. In this update, a contextual menu item (available by right-clicking or control-clicking) on a photo allows you to tell iPhoto to rescan the picture and detect missing faces, being less stringent with its recognition algorithm. The result is that Faces will be more likely to detect, well, faces.
The feature is a big improvement. There were easily dozens, if not hundreds, of photos in my iPhoto library where one or more faces weren't recognized when I first tried out iPhoto. One important point is that this feature works on a photo-by-photo basis, meaning that you can use it when a face is not recognized. Doing so, however, doesn't adjust the algorithm used to scan your entire library, which could result in extra hits of nonface objects being accidentally detected as faces.
In a similar vein, if iPhoto initially identified something as a face -- and you used the "x" icon in the face box to tell the software that the object was not actually a face -- rescanning those photos for missing faces takes you back to square one: iPhoto will re-identify the objects as faces.
Recognize manually added faces
In addition to rescanning for missing faces, the original option of manually adding a missing face is still available. This option allows you to draw a box around a person's face and tag them when Faces doesn't recognize them. (Even with the new-and-improved rescanning option, there will be times when it still fails to identify a face.) While this allows you to tag people's faces with their names, in the original release of Faces, it didn't help iPhoto recognize that person in other photos. With this update, Apple has extended this option.
Now, when you manually add a missing face in a photo, Faces will rescan just the contents of the box that you draw around that person's face using less stringent guidelines than the default. This allows the application to learn from the faces you've added -- a powerful addition, both in terms of recognizing faces in general and in learning who people are in your library.
Tagging misidentified people
One of the best ways to train Faces to recognize people is to first tag a good sample of each person and then look at the Faces list for them. iPhoto won't display the photos you've tagged, but its guesses about photos that may contain that person. This view allows you to simply click once or twice on each photo to indicate whether these guesses actually are that person, making it easy to quickly identify someone and train iPhoto to recognize the person better.
While this tool isn't new, in the original release you could only confirm or reject iPhoto's guesses. Now you can use a contextual menu on one of iPhoto's incorrect guesses to do more than simply reject the selection; you can also correctly tag the photo as someone else. It might sound like a small change, but it's a huge improvement when it comes to quickly training iPhoto to recognize people.