In this competition, we would like to tackle the problem of fine-grained comic character classification which lies between characters detection and identification of individual character (character’s name). The competition task aims at classifying the character’s images into one of the existing classes of individual comic characters. For example, all image instances of the character “Batman” should be classified in one same class “Batman”, all instances of the character “Daredevil” should be in class “Daredevil”. The visual distinctions between characters are often subtle, especially for characters in the same album, or in multiple albums of the same authors or the same style. Hence, it is difficult to address with today’s general-purpose object classification methods.

The comic book image analysis has been initially studied by the Document Image Analysis community. After that, it has been studied in the domain of computer vision using machine learning and graph recognition. The problem of fine-grained classification of comic characters could be handled by different approaches by different techniques. We encourage the participants to image-based approaches, graph-based approaches or hybrid-approaches.

  1. Image-based approaches (challenge 1)

The image-based approaches focus on the design of visual features (hand-crafted or deep features) then using heuristic rules or machine learning methods such as neural networks, decision tree or SVM, etc. to classify the comic characters.

  1. Graph-based approaches (challenge 2)

Graphs are a powerful relational data structure, which are very widely used in pattern recognition. Graphs provide an expressive and convenient way to represent the structure, topology, geometry and attributes of underlying information which has its applications in various domains including (but not limited to) computer vision, image analysis, data mining, pattern recognition and machine learning. Graph representation of comics has lots of relational information in general and the comic characters in particular exhibit challenging diversity in their shape and attributes while maintaining enough discriminatory information that permits their identification and recognition. This makes it an interesting challenge for graph matching, graph isomorphism, subgraph matching, subgraph isomorphism and graph classification to do fine-grained classification of comic characters.