Facing the Faces: Getting better results from iPhoto

David Braue
7 May, 2009
View more articles fromthe author

I’ve already written about my love of Faces, the iPhoto ’09 feature that automatically picks out the faces of people in your photos. The tool is great value for enthusiastic photographers who want to better keep track of the images they shoot, yet it can also be a frustrating pain in large photo libraries with thousands of images to tag.

For some people, these design flaws were a fatal problem that left the feature as nothing more than a novelty to be largely ignored. However, Apple’s point-point updates to iPhoto ’09 have improved several aspects of Faces considerably – and I’ve been making great use of them to plow through my photo library, which is currently hovering just under 34,000 images. To this end, I thought it might be useful to share some more of my experiences using Faces, and how I’ve learned to push it a bit to increase its usefulness.

Name that face. One of the biggest problems with the initial Faces release was that it would present a grid of faces, but only let you identify whether a given picture was or was not the person in question. Although you might recognise a face onscreen, you couldn’t tell the program who that face belonged to. This was a major problem not only because it wasted time, but because clicking on the person whose face that image belonged to, wouldn’t necessarily bring up that image.

Apple has since improved the face recognition algorithm so it’s a bit more forgiving, and I’ll explain in a moment why this is a good thing. When it comes to the naming grid, however, the biggest improvement is that you can now right-click on a picture in question and choose ‘Name’ to specify which person the face belongs to. It takes a bit longer than a simple yes-no scan, but it is far more useful in quickly whittling down your pool of unnamed pictures.

Another way of speeding through the Confirm Name screen, I have found, is to take a reverse approach to that which I suggested previously: since after the first few rows most of your pictures will be negatives, command-select each screenful of pictures so they are all red-tagged. Then scan through the images and identify which faces are matches, and which relate to someone else.

Measure your task. If you want to know just how many faces you have left to name, the updated iPhoto offers a way. Create a smart album with the parameter that ‘Face’ ‘is’ ‘unnamed’ and you’ll get an album with all the photos you haven’t named yet (and a count of how many meet this criterion).

You can, if you like, scroll to the top and click ‘Name’, then step through each picture typing in names; click on one name, type in the person’s name, and TAB to go to the next one. It’s time-consuming, but invaluable as there will always be faces iPhoto just can’t detect. My counter currently reads 10,479 – which is both a progress marker and a stress inducer. Sort of (more later).

Iterate like you mean it. I mentioned that Apple had improved Faces’ accuracy in this version. This does seem to make it more accurate in suggesting names for many faces, but it is far from perfect: many faces don’t pick up correctly even when they’re quite clear.

If Faces has already scanned your photos, the new algorithm is worth applying to your images. Apple has made this easy with a new right-click option called ‘Detect Missing Faces’; this goes through the selected photos and rescans the images using the new algorithm. Because I have so many images, I did it in groups: I searched for 2009, then did a Select All and right-clicked to do Detect Missing Faces. I then did the same for 2008, 2007, etc.

Rescanning can take a few hours (you now know it’s done when the spinning circle icon next to ‘Faces’ disappears), but it’s worth it: my unnamed image counter was down to 9200 unnamed faces, and after rescanning it had jumped to 11,029 unnamed faces. As I clicked through individual Faces profiles, this translated into the identification of a large number of images that Faces had previously missed – including over 100 pictures of me and up to 200 pictures of the most-photographed family members. These are images that would otherwise have languished in obscurity.

Boost your numbers – the smart way. Once the rescan is complete, click on Faces view and start with the first person on your panel. Step through the identification and tagging process several times, until you have picked all the photos of that person and identified all the other people with whom they’ve been confused. Then click the right arrow to go to the next Face, and start all over again.

When I initially started with Faces, it was my intention to only tag the faces of family members and close friends whose photos I would likely ever want to find. However, I soon realised that it’s invaluable to tag even people of passing importance – for example, work colleagues, school friends, teachers, and so on.

You may never want to pull up their photos by name, but during the process of naming their photos (from the Faces panel) Faces will throw up a large range of other possible matches – including, as I found, faces of family members and other people you actually do want to tag. Those faces may not a strong enough match for iPhoto to suggest them in the Faces view of the person you’re looking for, but they’ll appear as suggestions for other people. I found literally hundreds of images of people this way, even after the Faces views of those people were unable to throw up any more suggestions.

Tag ‘em and bag ‘em. One of the big questions many will ask about Faces is, ‘why go to all the effort?’ Reasons for tagging pictures will vary, but after recently purchasing my very first iPhoto picture book – and being very impressed with the results – I know Faces will be invaluable in helping track and group family photos for any number of future projects. I have several such projects going right now, including some of family members that may make their way under the Christmas tree (unless they’re reading this – in which case I’m really buying them a lava lamp!)

To make this process easier, I use Faces in conjunction with other keywords – for example, connoting weddings and so on. Keywords are also invaluable for photos where Faces just won’t pick up the faces, or where a person has their back turned but is still part of the photo.

There’s a trick to make keywording easier using Faces. You can’t use the Select All command on this screen, and it’s a waste of time to manually tag every picture. Instead, go to the Faces page of the person you’re interested in, then let it make all its suggestions in the bottom part of the screen. Click on the last picture that you have confirmed as belonging to that person, then scroll up to the top of the screen. Shift-click on the first picture to select all of the confirmed matching photos; use Command-K to bring up the Keywords window; and (carefully) click on the keyword for that person’s name.

Voila! You have both Faces tags and keywords in your pictures. That way, if you want to find a picture of your sister and her daughter at your cousin’s wedding, just type in their names along with ‘wedding’ and your chances of finding them are pretty good.

The unnamed masses. Although Faces is much better and more productive in its current version, Apple still needs to address one issue: the identification of swatches of fabric, bicycle wheels, food and other inanimate objects as faces.
While we all accept this technology is not perfect, these unnamed ‘faces’ chew up processing speed, inaccurately inflate the size of Faces’ database, slow down iPhoto’s processing, and – and this is most maddening – keep the Unnamed Pictures counter frustratingly high; even in my tagging heyday I couldn’t get it below 9200, but many of these were inanimate objects, grey squares (I’m still not sure what those mean) or people I will never ever want to tag.

Apple needs to offer a better way of dealing with these. For example, in the Confirm Name screen it would be great to be able to right-click on such a picture and get rid of the identified ‘face’. Perhaps there could also be a special area for unnamed faces of people that might be useful later on but should be excluded from future searches: I have hundreds of such faces, from people that keep popping up in my Confirm Name searches but for whom I will never add a separate Faces entry (remember that the more Faces you have identified, the slower the process works – so it’s a good idea to find a decent balance).

A few requests. Faces still isn’t perfect, but I’ve found the updated version can produce a lot better results If you work at Apple, or know someone who does, I have a few requests for the next version of iPhoto:

  • In the Confirm Name screen, add the option to Option-click on a number of faces, then identify them using the right-click and typing the name just once.
  • In the Confirm Name screen, add a right-click option to delete a face (especially useful when iPhoto has picked up stitching in fabric, wheels on a bicycle or any of a thousand other inanimate objects as being a face).
  • Force iPhoto to suggest at least one match for every photo it displays in Name view; in this case, a potentially correct guess is much better than no guess at all.
  • Offer a bucket to isolate identified faces of people that may be of interest, but who aren’t important enough to merit their own Faces entry.
  • Improve performance during manual naming: with such a large photo library, I can be waiting 20 to 30 seconds between clicking on an unnamed face and being able to enter the person’s name. [addendum: I should note that the AMW tip on speeding this process up by turning off item counts has done absolute wonders in speeding up this process, but Apple really could figure out a more efficient way of doing this by default. I like seeing the numbers!]

If you gave up on Faces early on as being too gimmicky or too inaccurate to be of value, I hope my experiences will help you get over that obstacle. With a bit of care and feeding, Faces can be an invaluable tool in keeping your photos under control.

Leave a Comment

Please keep your comments friendly on the topic.

Contact us