Sample images from the CAPD. Click to see larger version.

Examples of panels from (left to right, top to bottom) Flower, BNS, Words, Cats, Kim, A&B.. Click to see larger version.

About the CAPD

The Comics and Graphic Novel Panel Database (CAPD) is a collection of panel images organized by originating artist. Six artists, selected in a semi-principled fashion, contributed their work to the database. The CAPD also includes a control group consisting of twenty constructed images and over four hundred randomly selected images drawn from a search engine.

The following artists contributed their work to the creation of the CAPD:
    (A&B): J. Burgess, Able and Baker,
    (BNS): R. Perez & R. Coughler, ButterNutSquash
    (Cats): S. Ramsoomair, VG Cats
    (Flower): S. Notley, Bob the Angry Flower
    (Love): E. Kim, Love as a Foreign Language
    (Words): B. Rivers, Empty Words

All work is protected by copyright, and may not be modified, reprinted or otherwise used without permission. Rivers' work is part of the Creative Commons. CAPD samples have been scaled down in resolution and removed from original context, the interested reader is urged to see the artists' websites.

This database was collected in 2006 by T. Kowaliw at Concordia University in Montréal, Canada, and augmented in 2009 with a collection of images pulled randomly from a search engine. It is requested that any use of this database be accompanied artists' names, URLs, a statement regarding their retained copyright, and a reference to our paper and this webpage.


CAPD (zip; 34MB)
Just the documentation (PDF), including collection procedure.

CAPD Aacademic Usage

(If you do use the CAPD for academic work, please send me a copy of the paper and I will include a summary on this webpage.)

Kowaliw, McCormack, and Dorin used the CAPD in a classification task and to define a distance metric. They used evolutionary computation to evolve a collection of custom features, and then used a variety of nearest neighbour classifiers on that feature space for classification. They achieved, on binary classification tasks pitting the artists against all remaining images, a mean true positive rate of 0.944 and a mean false positive rate of 0.017. Further, they provide some evidence that a distance metric does correspond to naive human notions of similarity.
T. Kowaliw and J. McCormack and A. Dorin, ``Evolutionary Automated Recognition and Characterization of an Individual's Artistic Style'', in Proceedings of the IEEE Congress on Evolutionary Computation (CEC '10), 2010.

This page last modified Feb. 2010. Questions or comments may be sent to T. Kowaliw, taras DOT kowaliw AT ca.