ReVision: Automated classification, analysis and redesign of chart images
This paper[1] came to me by way of a friend and I am very gratefull for it. It discusses using computer vision to extract the data that is encoded into a visualization and the process of recreating a more perceptually effective visualiztion of the data. This is an interesting process in itself, but equally importantly it brought Cleveland and Mcgill's 1984 paper [2] to my attention.
I wonder if the world would be a better place if we were to wander around turning all the the pie charts into bar graphs. I am excited to see the renewed emphasis on quantifying the transfer of information rather than bragging about methods of mangling of information via dimension reduction.
I also wonder how this will hold up if results like the recent one that shows that bar charts with depth sometimes outperform flat bars. (find). Obviously, if that insight can be translated into some grammar than it can be put to use, but more generally I wonder if there aren't many features of natural visualizations which are lost in the process of normalizing to the Cleveland/Tufte approved standard.[1] M. Savva, N. Kong, A. Chhajta, L. Fei-Fei, M. Agrawala, and J. Heer, “ReVision: Automated classification, analysis and redesign of chart images,” in Proceedings of the 24th annual acm symposium on user interface software and technology, 2011, pp. 393–402 [Online]. Available: http://doi.acm.org/10.1145/2047196.2047247
[2] R. M. William S. Cleveland, “Graphical perception: Theory, experimentation, and application to the development of graphical methods,” Journal of the American Statistical Association, vol. 79, no. 387, pp. 531–554, 1984 [Online]. Available: http://www.jstor.org/stable/2288400
