(英) |
In recent years, it has been studied that various methods for detecting objects depicted in a comic by learning comic images. With conventional methods, however, the learning data used to detect objects in the comics is line drawings, and the amount of data sets for that purpose is limited. To solve this problem, this study examines how to detect dishes depicted in comics by learning actual cooking photos with a deep learning method. Our pro- posed method uses three types of cooking images such as (1) unprocessed (color), (2) gray scale, and (3) comic-style image. In our method, three types of cooking images were prepared: (1) unprocessed (color), (2) gray-scale, and (3) comic-like imaging. The objects in comics were detected by SSD300. This paper compared the accuracy of cooking image extraction in comics using this method. |