CNN-based Criteria for Classifying Artists by Illustration Style
CNN-based Criteria for Classifying Artists by Illustration Style
复制标题
基于 CNN 的按插画风格对艺术家进行分类的标准
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
Tatsuhito Hasegawa
中科院分区：
文献类型：
--
作者：
Kazuma Kondo;Tatsuhito Hasegawa
在艺术作品中，绘画风格作为艺术家的特征出现在图片中。这种风格可以用来识别假冒艺术品，而这项任务归根结底是识别图片的艺术家。艺术家分类被定义为图像分类任务的一种，因此适合使用在图像分类中取得高性能的卷积神经网络(CNN)来识别艺术家。然而，CNN的内部功能并不明确，也很难理解CNN如何识别图片的艺术家。有一种可能性是，CNN不是根据图片的风格来识别艺术家，而是根据其他标准。在本研究中，我们调查了进行端到端训练的CNN模型的标准。我们澄清了CNN根据插图的风格对艺术家进行分类。
In works of art, the drawing style appears in pictures as an artist feature. The style can be applied to identify fake works of art, and this task comes down to identifying the artists of the pictures. Artist classification is defined as a type of image classification task; therefore, it is suitable to use a convolutional neural network (CNN) which achieves high performance in image classification to identify an artist. However, the internal function of the CNN is not clear, and it is difficult to understand how a CNN identifies the artists of pictures. There is a possibility that a CNN identifies artists by not the style of the pictures but by other criteria. In this study, we investigated criteria of the CNN model which is performed end-to-end training. We clarified that a CNN classifies artists using the style of the illustration.