Calligraphy Word Style Recognition by KNN Based Feature Library Filtering

Chinese calligraphy works is a valuable part of the Chinese culture heritage. More and more calligraphy works are digitized, preserved and exhibited in digital library so that people can enjoy the calligraphers’ works conveniently. There are five main writing style categories of calligraphy words, namely, seal script, clerical script, standard script, semi-cursive script and cursive script. Users always want to appreciate the style-similar works simultaneously, so it’s necessary to classify the words by their writing style. In this study, we proposed a method based on KNN and feature vector filtering. Firstly, extract the SIFT points from the training images, building up a feature library for SIFT feature vectors. Then use a KNN-based method to filter the feature library so that the style-irrelevant feature points can be wiped out. At last we use the filtered feature library to classify the word images by a modified KNN classifier. Experiments show that SIFT feature has better recognition result than that of Gabor feature and GIST feature, but the large amount of feature vectors in the SIFT feature library makes the KNN searching rather slow. To accelerate the recognition speed, Spectral Hashing is used to index the feature library, which makes it faster to classify feature points and gives no side effect on the recognition ratio.