Inscription Detection and Style Identification in Chinese Painting

摘要

Chinese painting has a long history and enjoy a widespread reputation, the inscription plays an important role in the process of authenticating Chinese paintings, which includes the personal style. Previous approaches for scene text detection pursued the high accuracy based on the deep learning, but they rarely used calligraphy and painting works as the recognition target. It’s necessary for us to adopt new methods because calligraphy and painting have their particularities. In our work, we use the deep neural network model to extract the inscription of the paintings, and combines the characteristics of Chinese paintings to improve it. In the established data set, the model has achieved a high accuracy of 89%. Furthermore, the existing models focus on the recognition of the detection area, but style identification of the detecting area is our research focus. We use the style loss function to compared different painter’s style, and prove that the model has a promising performance for distinguishing styles. Our work not only meets the classification requirements for calligraphy and painting works, but also provides technical support for the identification.