Calligraphy plays a decisive role in the strategy of national culture.Calligraphy art information provides an important foundation for the protection and related research.However,the existing methods are difficult to extract the spiritual information of calligraphic accurately.The reason is that calligraphic strokes have complex variation in the ink,blurred edges and noise.In addition,calligraphic education faces by the serious problem of shortage of teachers.To solve these problems,we employ computer vision to help calligraphy learners to master brush movements,the main challenges are the writing gesture recognition and the brush tracking.To solve those challenges,the research content is as follows:1.Artistic features extraction from Chinese calligraphy works via regional guided filterTo extract calligraphy spirit accurately,we propose an extraction method based on regional-guided filter.It uses KNN matting to extract the basic information.Then,We filter the internal area and the edge through different filter windows.The algorithm can accurately extract the ink variation in strokes,and can well maintain the edge details of the calligraphy characters.2.Calligraphic imitation evaluation based on depth video analysis1)Writing gesture recognition algorithm based on MCNN-LSTMFor the problem of writing gesture recognition in the existing methods,we proposes an algorithm based on MCNN-LSTM.It proposed the multiple convolutional neural network(MCNN)to extract the spatial features of the video frame effectively,and is combined with the LSTM to more accurately extract the fine-grained motion characteristics during brush motion.2)Calligraphy imitation evaluation methodWe use Faster R-CNN and KCF to track the brush's movement trajectory accurately.Then we calculate differences between learner's brush trajectory.Finally we use the regression model to complete the prediction score.The error between the prediction score and the manual score are kept within ±3,which can provide effective guidance of calligraphy imitation. |