Feature Fusion based Cross-modal Retrieval for Traditional Chinese Painting

Recently, large-scale digital preservation of traditional Chinese paintings has been launched. Effective retrieving them is still a significant challenge. There are two main reasons: (1) compared with modern images, their color contrast is not sharp enough, and the realism of objects is different. (2) the lack of labeled datasets. In order to address these problems, in this paper, we build a new labeled dataset and employ cross-modal retrieval for images of traditional Chinese painting. Our method utility data in two modals, text and image. We use the Bert model to extract the semantic text feature, and use CNN to extract the image feature. The two types of features are fused to realize cross-modal retrieval task. We evaluate our approach on three datasets. Experimental results prove the effectiveness of our approach.