For the T-Sne grid I took a small subset of Teenie Harris photos and ran the program through them. One thing the T-Sne grid highlighted was the gradation in the number of people teenie put into a photo and the arrangements of the people. This is what lead to my ideas for my visualization project.
For my visualization project I decided to work on the Teenie Harris Archive of photos. One thing that was really important to me and museum was getting into Mr. Harris’ head. Figuring out how he composed he shots and patterns that emerged through that. My initial idea was that I wanted the angle of him/his camera towards his subject. However detecting angles and orientation of people are a very challenging and runtime expensive option. So i decided on distance and went with that. While the distance was able to be measured there are a lot of assumptions that had to be made with his camera, height, focal lens, etc. So while I got it to work currently I am working on a formula for calculating the margin of error so that people can know these aren’t precise estimates. Another thing I did with the archive is run face detection over all of the images to give the horizontal and vertical percentiles of all the faces that could be automatically detected. The horizontal and vertical percentiles of faces as well as the distances get outputted into JSON data. For my final project I decided to plot the average percentiles on top of the photos or to possible lay a semi-transparent heat map over the images to convey possibly where Teenie usually puts people in photographs to where he put them in each picture. This will involve things like variance, standard deviation and it depends on the distributions of the percentiles. The first thing I want to do is use D3 to make a vertical histogram and a horizontal histogram. Then from that information manually construct a heat map in the form of a matrix to overlay. (Higher numbers in the spot indicate that’s a common place for faces). Another thing I would like to do is run lines of best fit through people’s faces in group shots to get a sort of horizon line. And my most ambitious idea is in photos of only one or two people…. I would like to run a line of best fit from head to toe through each person so I could see how the photo is broken up. I think a lot of these lines would do well to be compared against the rule of thirds and hopefully by the end of this project i will have a new rule of Teenie. If i have more time then I’d like to create an interface to see all of this information.
All of my images can be found on github at :
My video can be found here: