Style Feature And Fisher Vector Based Classification Of Chinese Traditional Woodcut Paintings

Chinese woodcut painting can be traced back to ancient times.It is an important form of the Chinese nation traditional cultural expressions,a gem of Chinese forklore art,and is a precious cultural heritage that we should cherish and protect.At the same time,content-based image retrieval and classification has always been a hot topic attracting more and more attention in the field of image processing.The technique avoids the tedious tagging work of the traditional image retrieval and classification methods,which are based on annotations of words.However,the techniques try to retrieve and classify images according to their visual similarity,and to realize the targets of retrieval by images and classification by images.In recent years,a large quantity of new ideas and technical improvement are constantly proposed to this topic research.We creatively apply image retrieval and classification techniques to Chinese woodcut painting pictures,which have their own unique properties compared with other general images.Color,texture,stroke thickness and some other image features not only make this kind of images different from usual images,but also lead to their internal differences in age,origin and style.So our work not only puts forward a new attempt and possibility to image retrieval and classification technology,but also has made a small contribution to the treasure and protection of Chinese traditional culture.According to the style characteristics of Chinese woodcut painting pictures,we extracted the image features such as color,gradient,texture,brush strokes’ width,trying to capture the area style of the image and realize the classification by the image style rather than the content.And in terms of image classification,Fisher Vector framework is currently the most advanced coding technique of local features,so we also used this technique to encode a number of local sift descriptors into a high dimensional feature vector.This kind of encoding strategy has not only a high efficiency but also a low accuracy loss.Even a linear classifier is used,it can also produce excellent results.In our experiments,we adopt both the two image representation methods to the classification task of Chinese traditional woodcut painting images,and try to compare and analyze the classification results.