One million manga pages
Jeremy Douglass, William Huber, Lev Manovich. "Understanding scanlation: how to read one million fan-translated manga pages." In Image and Narrative, winter 2011. [pdf 3 MB].
Lev Manovich. "How to Compare One Million Images?" In David Berry, ed., Understanding Digital Humanities (Palgrave, 2012).
Keywords: manga, comics, Naruto, visualization, cultural analytics, softwarestudies.com, softeware studies, digital humanities, Calit2, UCSD, NERSC, HIPerSpace, Cultural Analytics, Lev Manovich, Jeremy Douglass, William Huber
"The humanities with some heavy iron...compared to other scholarly attempts to analyze Japanese comics — well, *gasp, choke, Good Lord!* Lookah that thing! It’s like some terrifying splash panel from vintage EC comics." Bruce Sterling, a blog post about our Manga visualization (see below), November 14, 2010, wired.com.
Using a unique 287 megapixel HIPerSpace at Calit2 (San Diego) for Manga research.
As soon as new Manga books are published in Japan, fans buy them, scan the pages, translate the text into other languages, and distribute digital images of the translated pages via web sites. In the process, they also insert additional pages (group credits, commentaries, and original fan art). This process is referred to as “scanlation.” Until July 2010, the most popular online archive of the “scanlations” was OneManga.com. (It was also among most visited web sites in general - 300th in the U.S., and in the top 20 in Singapore and Malaysia.)
In the Fall 2009, we have downloaded 883 Manga series containing 1,074,790 unique pages from this site. We then used our custom software system running on a supercomputer at National Department of Energy Research Center (NERSC) to analyze visual features of these pages (funded by Humanities High Performance Award from NEH Digital Humanities Office.) To match the scale of the data, we are using 287 megapixel The Highly Interactive Parallelized Display Space (HIPerSpace) with a a custom software for interactive exploration of large image sets developed between our lab (softwarestudies.com) and Gravity lab at Calit2.
Visualization below shows our complete data set - 1 million Manga pages organized in 2D space according to their visual characteristics. The pages in the bottom part of the visualization are the most graphic (they have the least amount of detail). The pages in the upper right have lots of detail and texture. The pages with the highest contrast are on the right, while pages with the least contrast are on the left. In between these four extremes, we find every possible stylistic variation.
(Note: some of the pages - such as all covers - are in color. In order to be able to fit all pages into the largest possible image our software could render and save, we made this rendering in grey scale. Also note that because pages are rendered on top of each other, you don't actually see 1 million of distinct pages - the visualization shows a distribution of all pages with typical examples appearing on the top.)
Manga style space.
Data: 883 Manga series from the scanlation site OneManga.com. Total number of pages: 1,074,790.
Mapping X axis: A mean of standard deviation of pixels’ grayscale values in a page. Y axis: A mean of entropy measured over all pixels’ grayscale values in a page.
What do we learn from this visualization? It suggests that the very concept of style as it is normally employed becomes problematic then we start analyzing large cultural data sets. The concept assumes that we can partition a set of works into a small number of discrete categories. However, if we find a very large number of variations with very small differences between them (such as in this case of 1 million Manga pages), it is no longer possible to talk about "style" of these works. Instead, it is better to use visualization and/or mathematical models to describe the space of possible and realized variations.
What about single manga titles? is it meaningful to talk about a title's style? As we found out, often even a short title has such graphic variability that we also can't use "style" concept. Here is an example of such title ("Anatolia Story"). 879 pages are organized by brightness mean (X) and entropy (Y):
In these examples, manga pages are organized according to particular visual features. Taking into account other features and also higher-order attributes (content, composition, manga's visual conventions for rendering characters, their faces, backgrounds, etc.) may reveal the presence of distinct stylistic styles in the one million pages sample, and also show more stylistic coherence in individual manga titles. We are hoping to investigate these questions in near future.
You can find a discussion of the initial results of our work with Manga data set in this research report:
Style Space: Analysis and Visualization of 1 million Manga pages (06/2010). [key 56.6 MB]. [ppt 17.9 MB]
More manga and comics visualizations created by Software Studies Initiative (Flickr)
Daniel H. Pink. Japan, Ink: Inside the Manga Industrial Complex, Wired 10.22.07.