My research is in the area of Document Image Analysis. To be specific, the topic of my thesis is to automatically recognize and index characters in a set of hand-drawn objects, e.g. given a volume of comics --> how to localize, recognize then index the characters in it (e.g. return "Lucky Luke" position in some random volume).

First thing first, before I can work on the indexing part, I must have some reliable system to recognize the interested characters. After some months working on it, I find the recognition is extremely hard, because the drawings are some time very "abstract".

For normal face recognition, there are a lot of researches on it, thus we have Haar-like features, biometrics features, etc. For normal photograph objects, we may use e.g. SIFT/SURF, or other strong features for the purpose of object recognition.

But in cartoons or comics, we may extract so few or zero feature(s) following the traditional approaches. Human visual interpretation system is amazingly effective in recognize such abstract objects (e.g. only with 4 lines and 2 curves, we still recognize the familiar character).

So I think the solution must be somewhere in the understanding how human can interpret such a set of line strokes as a meaningful object. Once we understand it we can try to "teach" the computer the same way.

Thus are there some good approaches in the terms of recognition of abstract objects such as drawn comic objects? (I searched for some but they're not so relevant). Thanks.


migrated from cstheory.stackexchange.com Apr 10 '14 at 20:51

This question came from our site for theoretical computer scientists and researchers in related fields.

  • $\begingroup$ Thanks for your suggestion. I was afraid that I posted the question in the wrong site, but from this site's description: "Theoretical Computer Science - Stack Exchange is a question and answer site for professional researchers in theoretical computer science and related fields. For undergraduate-level questions please visit Computer Science which has a broader scope." I don't think the problem introduced here is related to some undergrad- level of CS. $\endgroup$ – Jim Raynor Apr 10 '14 at 11:13
  • $\begingroup$ I suspect the downvotes and close votes are because of the vagueness of the question and the unclear connection to theory, and not a statement about the research-level-ness of the post. If you want it migrated to CS.SE then flag it to let me know. $\endgroup$ – Artem Kaznatcheev Apr 10 '14 at 19:41
  • 1
    $\begingroup$ Consider researching what techniques cartoonists use to make a character recognizable. As anybody who has played Pictionary can attest, not everyone can draw an object and make it immediately recognizable. $\endgroup$ – Dan Bryant Apr 15 '14 at 18:37

It sounds like you're looking for shape matching techniques. Sven Loncaric's well-cited survey paper "A Survey of Shape Analysis Techniques" from 1998 can be found here.

In 2002, Belongie, Malik and Puzicha proposed a similarity measure between two shapes based on

  1. solving for correspondences between points on the two shapes, then
  2. using the correspondences to estimate an aligning transform.

David Marr's book 'Vision', particularly §3.6 and chapter 5, also provides some background about the correspondence between shape recognition in human visual systems and computers.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.