Wrinkled Chinese Handwriting-Based Writer Identification

Targeting at the problem of wrinkled Chinese handwriting recognition, this paper presents a method based on statistical characteristics of scattering transform coefficients. This method mainly uses the local translational invariance and the stability of elastic deformation of scattering transform. Firstly, the text image is transformed by scattering. The distribution characteristics used as global characteristics are extracted by applying gamma model to scattering coefficients of sub-bands. Then the copula model is constructed based on these global characteristics. Finally, the similarities are calculated for recognizing writers by using the Kullback-Leibler divergence between copula models. Theoretical analysis and comparative experiment show that our method based on statistical characteristics of scattering transform is more advantageous than the others for when regarding text images with various degrees of wrinkles.