The Research Of Seal And Inscription Localization In Chinese Painting And Calligraphy Based On Convolutional Neural Network

Chinese painting and calligraphy is the precious treasure for Chinese culture in the past long history.Digitalization for Chinese painting and calligraphy has grown into a trend due to its convenience in reservation and Dissemination.Due to the difference with common scene image,traditional image processing methods are not suitable for Chinese paintings and calligraphy.As an important part of Chinese paintings and calligraphy,seal and inscription consist of great semantic information.Detection of seal and inscription is a fundamental research in this field and is helpful for Chinese painting and calligraphy retrieval.In this paper,a method based on convolutional neural network(CNN)is proposed to achieve accurate seal and inscription detection in Chinese painting and calligraphy.The proposed method combines priori knowledge in the scene and CNN detector.Color and edge features are used to extract candidates,then accurate object position is gained using pre-trained CNN detector.CNN can learn high-level features from large amounts of training examples,which are more natural.In the proposed method,CNN models help to acquire high detection accuracy and candidate extraction helps to improve efficiency.Furthermore,hierarchical clustering in the inscription detection method compensates for some missing text regions and rejects some false positive regions,improving accuracy of the method.In this paper,a seal dataset of 15571 samples and an inscription dataset of 15448 samples are constructed.Experiments in 694 Chinese paintings and calligraphy indicate that the proposed seal detection method outperforms traditional seal detection method and achieves high accuracy.Moreover,the proposed inscription detection method is proved to be feasible and efficient in experiments of 484 Chinese paintings.