Multi-feature Fusion Based Classification Of Chinese Woodblock Paintings

Woodblock New Year painting is an old folk art form originated in China, and isa valuable cultural and historical heritage. During its history, Woodblock New Yearpainting has been a popular and necessary kind of painting, which combines blockprinting, painting art and folk literature together. As the essence of Chinese cultureand art, Woodblock New Year painting has been researched long before.With the time develops, people’s daily life has greatly changed, a great deal ofcustoms in Chinese Spring Festival have been forgetting gradually. Each school’sNew Year painting has disappeared one by one. Jicai Feng, the president of Chinafolk artist association, leaded a team to carry on an unprecedented and large-scalesurvey and restoration. During the survey, it is easily to find that many imagesinformation about schools and subjects are hardly to be verified. When art workerswant to do some research on the paintings, the shortage of information will cause a lotof trouble. So verifying the information is extremely urgent. The recent advances inmachine learning and multimedia feature extraction have made this task easier toautomate.In this paper, image classification technology is applied to classify WoodblockNew Year paintings based on the product area and theme. The classification enrichesthe text data stored in the database. Image processing and pattern recognition are usedin the pre-processing and feature extracting of New Year paintings. Feature Fusionbased on Multi-kernel learning is the key method in the classification of the paintings.In addition to the general features, two specific features for Woodblock New YearPaintings, Saturation entropy histogram and Region area histogram, are found.Multi-kernel learning improves the accuracy of classification comparing withSimple-kernel learning, which is shown in the results of our experiment. Furthermore,a system for the prediction of the Woodblock New Year paintings information isdeveloped.