Teenie Harris was a leading African-American photographer and photojournalist, who was active in Pittsburgh from the ~1940s-1970s. Through my internship at the Carnegie Museum of Art, I’ve been lucky to have access to the museum’s large database of Teenie’s photographs.
This book is catalog of all the faces in a (small but representative) sample of 2000 photos from the photographer Teenie Harris. I used the dataset of photos and a JSON output of a facial detector to isolate all the faces in each image; compute the brightness of each face photo; and sort the faces by brightness. The effect was a book that goes from dark to light, eventually fading into the background color of the page.
Most of my process was oriented around creating the code to generate the book. At first I needed to figure out how to retrieve all of the pixels that contained the face, so i used the bounding box that the face detector’s JSON output gave me. Then for each set of pixels I created a face object with the original size, the pixels, and the brightness. When creating the new face object I had to resize each face to a standard 20 pixels by 20 pixels. Then I sorted each face object by brightness. Then I looped through the sorted faces to create grids of faces that looks like this:
What I discovered was that the charm of the book wasn’t really the gradient but all of the variation within the gradient: the hidden pockets of light and shadow, the variation in face orientation, the expressions that deviate from the traditional portrait smile, and the occasional (and very rare), misclassified “face”. My inspiration for doing stemmed from the practice of of african americans lightening their photos to look “light-skin” or “passing white”. At first I wanted to see if I could catch any hint of that through sorting the faces, but then I realized i had no way of knowing the actual skin tones of the people in the photo versus the exposure that he used.
All things considered, I think this was a successful project. If i were to change anything, i’d probably make the book double-sided instead of flip book style. Additionally I would run the script on the full dataset of images (which i did not have at the time) instead of a sample. Also I struggled a lot with sorting because I couldn’t figure out how to bring in built in java libraries, and so i implemented quicksort which gave me a lot of bugs, so in retrospect I would’ve just asked someone for help sooner.