Leveling the Color Field:
A Tool for Entropic Image Manipulation

Overview

The probability of color distribution within an image can be calculated based on the total color values in the image. We use an iterative process to shift these values toward an entropic state, with a slight bit of randomness involved in the redistribution. Changes are more radical toward the outside edges of the image, so that the center remains largely untouched. The edges and objects within the image are still discernible; their color patterns, however, decompose and change to achieve the desired color balance.

Background

The piece is a response to a project called "Background Amplification" by Ben Fry which expands the borders of an image by adding similar pixels. We wanted to explore the concept of entropy by doing the opposite: deconstructing an image from the outside in. In doing so, we hoped to combine the concept of entropy as defined by the second law of thermodynamics (systems naturally move toward states of greater unuseable energy) with the statistical definition of entropy (a measure of disorganization) in terms of color. We wished to constantly move an image toward a state of color disorganization (i.e. less ordered distribution) while maintining the image's patterns.

Implementation

A user loads an image into the application, which scans each pixel to find four values: red, green, and blue (as 8-bit color values), and distance from center. It then counts the number of instances of each color value, compares these numbers against the total value for each color, and thus determines the probability of each value existing in the image. The standard entropic calculation (using summation of probabilities) reveals the "Color Entropy" for each of the three hues (red, green, and blue); these are compared against the total strength of each hue in the image. Next, an iterative process goes through the image and alters each pixel to slightly increase the evenness of color distribution toward a centered state. Each pixel is compared against the mean, and shifted toward it based on a weighting strength that is proportional to the pixel's distance from the center of the image. After all pixels have been slightly altered, the process recalculates the color entropies and total strengths. The iterations continue until the image has been effectively re-rendered in an entropic state.