Generating true color paper textures of historical documents

This paper presents two methods for automatic generation of true color paper textures of historical documents. One uses a 256-gray level image of the texture, some features inherent to the original paper itself, and neural networks to colorize. The other generates a synthetic version of the texture based on the rebuilding of its histogram. Both methods produce images which are perceptually close to the originals by visual inspection and quantitatively by the use of variance analysis. With the colorizing process, it is possible to create a cluster of textures and use it to generate different textures from just one sample. The synthesis process allows the complete generation of the image of the document achieving high compression rates.