Tesserae: Google Image Art
Tesserae
After the critique last week I decided to change my topic from mapping the colors of the tree of life to comparing the meaning of words in context. This idea came from the realization that images of closely-related species take on different colors based on whether that species lives in captivity or in the wild. One chimpanzee had a green hue in the visualization whereas its close siblings were all yellow. It turned out that the difference was due to green leaves in the first chimp’s environment and the yellow zoo-bedding in its siblings’ environments.
My final visualization (named Tesserae, the latin word for tiles) was created in order to display these differences visually and allow people to test hypotheses about the contextual meanings of words like the names of the two chimps. It uses an HTML interface that allows users to type in related words and see their visualizations side-by-side.
One frustration I had with the previous project was that the color average used to color the nodes lost the rich textures of the underlying images, leading to an uninteresting visualization. To remedy that problem I took 15×15 pixel chunks out of the source images, preserving the textures while obscuring the overall image. This allows the visualization to become an composite that abstracts away the meaning of the individual pieces, making the user to focus on the relationships between images rather than the images themselves.
Since all of the images are downloaded and client-side, no two visualizations are the same. The tiles are arranged randomly on load and the images themselves change as Google’s search index updates.
Analysis
In practice Tesserae is more useful for aesthetics than for practical comparisons between words. I’ve been using it more often as a story-telling tool than a data analysis one. This visualization is a good example. It uses the search terms “stars,” “earth,” and “sun” to paint a picture of our solar system.
Some meaningful comparisons do exist, however. They just seem harder to find. One that I found a little interesting was the difference between the names “Alex” and “Alexandra,” shown below. Tesserae does all image searches with SafeSearch off, so the image for Alexandra is full of pink-tinted pornography. Alex has no porn and is a washed out gray. If I put on my “culture” hat for a second, this might say something about how the internet views women and men differently.
There are a few more examples worth looking at on the project’s website.
Room for Improvement
Aside from the performance issues listed on the project website, I can think of a few places where the visualization could improve:
Since it only grabs the top eight images from Google Images it’s easy to pick different images out of the composite. A larger number of search results might be beneficial but is cumbersome to implement using Google’s search API.
A save button could be added to make it easier to share generated visualizations with friends.
The original photos that created the mosaic could be displayed on mouseover. This would make it easier to find out why a mosaic turned an unexpected color.