In my last blog-post, I touched on a potential method to study neuroesthetics by first analyzing the specific characteristics of some artwork using Adobe Photoshop. While I have continued to look at possible ways to quantify artwork characteristics, I first decided on a methodology. I plan on using Amazon’s Mechanical Turk, an online-based method designed to get human subject input. I can create a dataset of artwork (discussed later) and have each subject judge which artwork he or she finds more pleasing from a random selection of two. From this, I can gather data on which artworks are preferred more, and relate that to the characteristics of the artworks. For example, if as the contrast in artwork increases, the preference also increases, then perhaps that is an interesting correlation worth studying more.
For creation of the dataset, in my last blog-post, I considered Abstract Expressionism. After discussions with Dr. Alison Langmead and Lily Brewer, I have come across several other possible movements of art that could be used to create my dataset. What is important is that this dataset is not a random collection of artwork from different time periods and different artistic movements. To reduce potential bias from subjects simply preferring one style of art over another, they should all be somewhat similar in theme (or even all from the same artist), unless one wants to compare two different styles. For example, Minimalism (Agnes Martin, Frank Stella, etc.) was a movement that partly originated as a reaction to Abstract Expressionism, so it may be interesting to look at both minimalist and abstract expressionist artworks together. Separate from the Mechanical Turk project, analyzing these movements could show more similarities than previously thought (e.g. maybe they differ in shapes and lines but use color and contrast similarly). Another potential robust genre of artwork for the dataset is landscape paintings (Hudson River School, Dutch Landscape Painting, etc.), as many of them have a similar theme but may differ in qualities like contrast and color. Here, another potential comparison would be landscape paintings versus landscape photography (Timothy O’Sullivan, Ansel Adams, etc.). Currently, I am considering all of these movements/genres and analyzing few works from each, and will have selected one (or two for comparison) for the dataset by the end of this semester.
In analyzing these few works, I have been using Adobe Photoshop. Photoshop can be used to find pixel measurements throughout the artwork, especially via the histogram feature. Below are the histograms of two minimalist artworks, Harran II by Frank Stella and Untitled by Agnes Martin.
From this window, the mean, standard deviation (std dev), and a median pixel value, all with a value from 0 to 255, can be found for a variety of channels (RGB, Red, Green, Blue, Luminosity). In addition, one can also find the Gray Value minimum, maximum, mean, and median of the work, which are measures of general brightness1. Below are tables with the histogram information for both works:
As far as comparisons between the two images, nothing within these measures is surprising; for example, there is certainly less color variation in Martin’s work, as seen in the lower Std. Dev. In my last blogpost, I considered a method to look at symmetry of artworks. After looking at these histograms, I reasoned that I could use the same method to look at symmetry again. This time, I would simply compare the histograms of the left and right sides to show possible similarities between the general characteristics of each half. Below is the data from those histograms:
As far as looking at the general pixel characteristics of both the left and right halves of each image, this method proves useful, though it does not say much about line or compositional symmetry. Both halves of Agnes Martin’s work have very similar colors (and are otherwise very symmetrical), which is shown by the close histogram values for the left and right side. However, in Stella’s work, the left and right halves use fairly different colors, and this difference is shown especially in the margin between the means/medians of the halves for all channels.
I think this method for analysis of the artwork is very promising for looking at color usage in these artworks. As far as the symmetry analysis goes, a further step could be to overlay the histograms of each half over one another and measure the common area. This would show more accurately how similar these histograms are, since they are not unimodal and the mean/median and standard deviation could potentially be similar for two very different histograms.
In my next blogpost, I will further discuss movements being consdier for the dataset as well as development in this Photoshop histogram analysis.