Dominant Colors - Data Visualization
The goal of this project was was to investigate the reduction of photographs to a single dominant color. In some sense this color is a summary of the photograph, but it is also, notably, a de-contextualization which doesn't retain any of the meaning of the original photograph.
I used a sample of 1382 photographs out of a total of 10,000 collected (200 photographs from each of the 50 most popular tags on flicker), selecting for pictures which had a location tag. I used Python and the modified medium cut method (for color reduction) to create the following data visualizations. To the right is an example of dominant color found using this method (photo by Flickr user Nigel Bewley).
After reducing each of these 1382 photographs to a single dominant color, I created an ordered list of these colors by their HSV color value (which considers hue first, then saturation, and finally value, or brightness) to visualize the frequency and variety of dominant colors.
This method of visualizing the dominant colors is more interesting when distilled by content tags assigned by the photo's uploader. Hover over the following examples to see what tag the dominate colors have in common.
Next I wanted to investigate where similar colors were grouped by physical location of the photograph's contents. (This seems reasonable if all the photographs were in the landscape tag or were all taken outside, but one should bear in mind that these photographs could be portraits, or of objects that could exist anywhere on Earth.) By plotting each dominant color as a dot (in that color) on a map, potential patterns may be revealed. Instead of eliminating black and white photographs, I opted to distinguish them with a square instead of a circle.
This map did not reveal as noticeable a pattern as I had hoped, so I pursued another method, a combination of the ordering above, and the location mapping. In the image below, I ordered the dominant colors not by value, but by latitude (from South to North). Latitudes are listed just above the colors, and some major cities where photographic dots were particularly dense are listed above that.
Photographs often hold meaning in their contents which cannot be determined readily from color alone; does color reduction then make the photograph meaningless? Perhaps color carries it's own symbolic meaning. A pattern cannot be determined from the map alone, and one is only slightly more apparent in the latitudinal ordering. However, some interesting things can be read from these data visualizations.
One may easily see that over twice as many photographs were taken and uploaded between 40 N and 50 N than are taken in the whole Southern hemisphere. One can also easily see bands of similar colors, but finding meaning in them is more difficult. The grey area around New York is readily explained; but other areas are just stripes of color, missing the content and subject of their photographs. For me this was an interesting study in color and an fascinating project to pursue. But more importantly, this photographic reduction serves as a reminder that some data visualizations are simply abstractions, requiring more narrative and context in order to derive much meaning.