In the top graphic, we can see the difference in the mean values of red and blue. This break plots the images based on their relative position, as this is an additive color mixing scheme, the higher values of each are most often found in white images. In this zoomed-in image, the distinctions between the images of white dresses against blue backgrounds break from the red dresses against non-white backgrounds.
The second large image is a study of the height of images. Many modern image sharing platforms are committed to the portrait page layout. This stands in contrast to the theories of new image technologies from the past that suppose that graphic design rules would adapt to layout as a landscape. Sharable social images are portrait.
These last two images show the propensity of the image set toward disorder and a zoomed out view of the color scheme. Each shows that the dominant region for these images is relatively small and consistent.
When we began the project, we predicted that we would see an average image of a woman in a white dress facing away from the camera. She would have sleeves and wavy hair. We thought that almost every model would be a white woman, and that almost every dress would have some sort of emblishment: lace, beading, jewels.
We found that the images were quite consistent with what we were expecting. There were a few important differences, Altough we did not count, many of the women, likely a majority were facing foward. The average image pinned of a wedding dress is remarkably consistent. the image is white, has low entropy, and likely very few details in the background.