Low-Quality Character Recognition Based on Dictionary Learning and Sparse Representation

Abstract

Dictionary Learning (DL) and Sparse Representation Classification (SRC) have shown great success in face recognition recently. Practice have proven that SRC has strong robustness against noise and occlusion in face images. Our work focused on a new low-quality character recognition method based on DL and Sparse Representation (SR). SRC is introduced to deal with the low quality of character images, such as broken stroke, noise, fuzziness. Simultaneously, we also apply the linear combination of over-complete dictionary to recognize characters with different fonts and sizes. A dictionary learning method based on factor analysis is also proposed to make the dictionary more discriminative. Experiments show our method not only can recognizes characters with different fonts and sizes, but also is robust against broken stroke, noise, and fuzziness. Our method is also efficacious as it does not acquire some complex preprocessing procedures, such as binarization and refinement.