I’m not sure what I think about Simon Høgsberg’s ‘The Grocery Store’ project,which I read about in a post on DVAPhoto. It’s certainly remarkable, made from around “97,000 photos of people outside a grocery store in Copenhagen” which were then analysed by the facial recognition algorithms in Picasa – freely downloadable photo software – to identify people who appeared in multiple images.
It’s worth reading the interview with Høgsberg by DVA’s M Scott Brauer which explores the how and why of the project and some of the issues, particularly around privacy involved, though I feel this could have been investigated more.
The images were made by Høgsberg “returning to the bike rail outside the supermarket with my camera” and zoom lens on 159 afternoons and “Freezing face after face with a click.” They certainly seem often to be very carefully chosen moments – as you can see from exploring some of the 2067 images that make up the web project – which is a very impressive one, with the images laid out on a single zoomable page as a grid “of sequences of images crossing each other in horizontal and vertical lines. Each sequence shows the same person caught on different days” and ” are arranged in chronological order.” It’s easier to look at than explain.
On Høgsberg’s web site there is more about how the project was carried out and his discovery of the face recognition in Google’s Picasa, software which enables you to “Organise, edit and share your photos” and share them with your friends on Google+L
Picasa uses facial recognition technology to find and group similar faces together across your entire collection of photos. By adding name tags to these groups of faces, new people albums are created.
The link tells you how it is done. Picasa is software I found rather annoying when I played briefly with it (here is a set of images of Paris I shared in 2006, complete with a multiple spelling mistake), but it seems perfect for this project.
Høgsberg gave some people in his images tags, starting with A1, A2, A3… and Picasa then sifted through the images to find the same people in other pictures. One man, E46, turned up in 276 of them. These sets were used to construct the project image.
There seems to me to be some theoretical problems here. Lets consider three people, who we could call A, B and C, and assume that there is a picture including A and B, another including B and C and a third including A and C. If the set of pictures of A is laid out horizontally then the set of pictures including B could be laid out vertically, with the picture including both at the crossing point. But if you then want to add the series including C, it can either be set to include the image together with A as a vertical, or that together with B as a horizontal series, but not both. And if A and B appear together at several different times, what then? Don’t even think about A, B and C all turning up at once…
Perhaps these kind of problems are why only around 2% of the images taken appear in the final presentation, though I imagine the interest and quality of the images was also a consideration.
But these are technical matters, and it is perhaps the privacy implications that concern me more. I wonder what ‘E46’, ‘R51’ and the others make of this project and their inclusion in it.
Its also a project that makes me think about the millions of images gathered every day by security cameras in various public places, and the kind of analysis and use to which they might be put – with the aid of far more powerful software tools than that included in Picasa.