Coloring in the Timeline of Art History
OverviewArt is a broad term that can encompass theater, music, literature, paintings, sculptures, and much more. Art has been created and preserved around the world throughout time. Within this project’s context, our team will focus on paintings from the Art Institute of Chicago. We want to understand the way that colors get represented within paintings geographically and chronologically. Over the next few weeks, we will create a time series visualization that correlates region and the painting’s dominant colors.
What is Art?
There have been many debates about how to define “art.” Often, art refers to some form of creative expression, be it painting, music, dance, theater, sculptures, quilting, and more. The term visual art often refers to art pieces that can be seen, including paintings, watercolors, pottery, sculptures, and electronic graphics. Since “art” can be an impossibly large category, for our analytics project, we will be studying paintings. The Eden Gallery defined painting as “the process of applying paint, or another medium, to a solid surface – usually a canvas.”
Who has preserved historical art?
When studying art, especially older artworks, it is important to remember that historians only have a sample of the art pieces from each time period. Many art pieces were lost due to age, natural disasters, wars, or political art destruction, while other art was lost simply because no one tried to preserve it. Four main groups have been most influential in funding and preserving art: wealthy individuals, educational institutions (i.e., universities, museums, galleries), governments/political leaders, and religious institutions. These institutions had huge power over what art was created, what art was deemed “good,” and who could create the art. For more information about preservation bias, read this FineWoodworking blog about how most preserved furniture was from the ultra-wealthy.
Why do we study art?
There are many reasons historians study art:
- Art can document historical events. Many paintings are scenes from wars, political leaders, revolutions, and other historical events.
- We can learn about the culture. Admiring statues of Poseidon can teach us about the Classical Greek worldview.
- Art can connect us with other people’s emotions. Seeing a painting of Paolo e Francesca can evoke memories of young love, or we can empathize with the anxiety of Munch’s The Scream.
- Art is beautiful and brings people joy. We can appreciate art for just being beautiful!
Our Project Goals
We want to understand the way that colors are represented within paintings geographically and chronologically. Artists used materials available to their time and region, often resulting in distinct color schemes and styles. For example, below are paintings from two different time periods. The Dutch artist Rembrandt’s Portrait of a Man in a Tall Hat was created around 1663 and exhibits the traditional dark colors associated with the Baroque period. On the right is the French painter Renoir’s work, The Dancer, which was created in 1874 and exhibits the bright colors popular in the Impressionist era.
Time Series Analysis
Using online pictures, we will use several data analytics techniques to determine the dominant colors in each painting and compare these colors by location over time. To do this, we will create a time series visualization that correlates region and color. Art pieces belonging to specific time periods will be grouped together and then further categorized by the region in which they originated. This way, we can analyze the trend of dominant and secondary colors of paintings.
Dominant Colors in Art
Dominant color in art is defined as the one that serves as a photo’s focal point. This color holds its hue despite the surroundings and will push through the painting. Color dominance can be established through strength of hue, sharpness, contrast, and perception of color.
In this project, we will attempt to locate dominant colors from a source image with OpenCV and K-means clustering. The large idea is to sample colors from a source image and build averages from clustered samples and return a best estimate of dominant colors. We are still streamlining
this approach and will have more details in the coming weeks.
The Data Set
We will use a subset of the data set from the Art Institute of Chicago. The Art Institute of Chicago (ARTIC) has over 122K artworks in their collection, sourced from around the world and spanning millennia of human history. In 2019/2020, ARTIC launched a public API to facilitate
access to their collection metadata.
For this project, we will filter the dataset for all paintings using the full text search feature that the API exposes. This enables us to get a subset of fields that are of interest. The entire dataset, not including the images, is around 1.75 GB. Since the paintings are nearly 4% of the full collection, we expect the dataset to be a lot smaller. A simple full text search of paintings indicates that we have around 3353 paintings in this archive.
Each painting has some fields that we think might be useful:
Readable description of the creator of this work. Includes artist names, nationality, and lifespan dates
Name of the preferred artist/culture associated with this work
The location where the creation, design, or production of the work took place, or the original location of the work
Dominant color of this artwork in HSL
Name of the curatorial department that this work belongs to
Unique identifier of the preferred image to use to represent this work
The images for the artwork are delivered via the IIIF Image API 2.0 which is a standard API to deliver structural and presentation information about digital content proposed by the International Interoperability Framework (IIIF) group. We will use this API to get the images for the artwork to analyze the dominant and non-dominant colors using our algorithms.
The Flaws of this Data Set
As previously discussed, there has been long-term, systemic bias as to what art is and is not preserved in museums. While ARTIC has a sizable, digitized art database, a group of people—with their own unique expertise, biases, and resources—selected which art pieces it would display, store, and digitize. Since the ARTIC set of paintings is not an unbiased, random sample of all art, we will not be able to definitively say our color timeline represents all art.
Additionally, the colors may not represent the original color of the painting. For older paintings, it is possible that the colors had faded over time. On the digital side, the colors of the digitized versions of these paintings may be slightly different than the original color of the painting. Depending on things such as lighting exposure can affect how the digital image turns out—and cause international debate, such as what color the dress is!
Despite the imperfect data set, we are still hoping to achieve the following things from this project:
Learn about the changes in dominant colors over time in the ARTIC’s collection.
- Establish which regions and time periods are most represented in ARTIC’s collection and suggest further studies for lesser represented regions/time periods.
- Contribute the use of machine learning to the study of art history.
- Open a discussion about the benefits and drawbacks of using algorithms to study digitized images.
Shivani Garg: My name is Shivani Garg. I did my undergraduate education at Virginia Tech where I double majored in Computer Science and Computational Modeling & Data Analytics. Since then, I have been working as a Data Scientist at Booz Allen Hamilton and recently started working at PNC Bank. What I love most about this field is how it can be applied to so many different industries and contexts. This is what makes me excited to embark on this project with my team. You can connect with Shivani on LinkedIn. Christine Horting: My name is Christine. I am completing my MS in Analytics from Georgia Tech in May of 2023. I have experience working as a music therapist in various mental healthcare facilities, and I transitioned into data analysis to contribute to researching and developing data-driven solutions to improve community health outcomes. J. Vishwanath: My Name is Vishwanath. I did my undergraduate degree in India where I majored in Computer Science and Engineering and then went on to get my master's in finance from Harvard University. I am currently working as a Quantitative Analyst, managing a quant team focusing on Environment Social and Governance factors and creating value-based investment portfolios.