Analysis of Chinese Calligraphy Fonts Using Content-based Image Retrieval

Abstract

Chinese have a very special category due to individual skills and font variations and present rich and varied visual forms, “calligraphy”. Traditionally, optical character recognition (OCR) for Chinese characters has achieved good results. However, calligraphic characters have more diverse fonts and backgrounds, which are also needed to be further recognized to learn and copy. Therefore, we establish an image retrieval system for calligraphic characters to present different styles and fonts, allowing users to learn from retrieved similar images. The user inputs the query calligraphy image to retrieve from the system. Then, the system automatically finds the calligraphic images with similar styles and the closest characteristics from the database for the user to copy. We use a calligraphic database provided by United Digital Publications, which contains each calligraphic style such as seal script, clerical script, cursive script, running script, regular script, and a total of about2,500 scanned images of calligraphic characters. It also contains metadata indicating the dynasty, age, and author of each image. In the experiment, a convolution neural network is used to build a classification model and extract the features of each image in the target image database. While querying, the system automatically calculates the cosine similarity between the features of all images in the target image database and the features of the query image. The top ten images with the highest similarity are used as the retrieval results. As a result, mean average precision (mAP) is used to evaluate the retrieval performance of the system. The experimental results show that the mAP of this system is 0.91, which indicates that the system can retrieve the calligraphic images of a specific calligraphic style effectively.