NMF-based approach to font classification to printed English alphabets for document image understanding

초록

This paper proposes an approach to font classification for document image understanding using non-negative matrix factorization (NMF). The basic idea of the proposed method is based on that the characteristics of each font are derived from parts of the individual characters in each font rather than holistic textures. Spatial localities, parts composing of font images, are automatically extracted using NMF. These parts are used as features representing each font. In the experimental results, the distribution of features and the appropriateness of use of the characteristics specifying each font are investigated. Add to that, the proposed method is compared with the method based on principal component analysis (PCA), in which various distance metrics are tested in the feature space. It expects that the proposed method will increase the performance of optical character recognition (OCR) systems or document indexing and retrieval systems if such systems adopt the proposed font classifier as a preprocessor.