Who Draw This? Caricaturist Recognition using Convolutional Neural Networks


Hicsonmez S., SAMET N., Akba E., DUYGULU ŞAHİN P.

26th IEEE Signal Processing and Communications Applications Conference (SIU), İzmir, Turkey, 2 - 05 May 2018 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/siu.2018.8404770
  • City: İzmir
  • Country: Turkey

Abstract

Caricature is the art of expressing an event that will be described in many pages in writing, visually with a small drawing. The details of the drawing, the style, and the objects used in the drawing, which make up the visual structure, are very important in conveying the event or thought. Therefore, different caricaturists express similar events (eg. social or political messages) using different details and objects according to their own style and imagination worlds. Determining which caricature belongs to which caricaturist will help us to organize large collections, determine originality, and make caricature suggestions based on subject/style. In this study, a total of 4212 caricature from 10 different Turkish caricaturists were collected from the Internet and a new caricature data set was created. Using this dataset, both the existing Convolutional Neural Network models and the proposed new network, ComicNet, are trained and their performances are compared. Experiments show that ComicNet is the most successful model with an accuracy of 94.68%.