A token-based file may be compressed by applying a medial-axis transformation to the tokens in the file. This generates a topological skeleton of the token that is a thin version of the token, equal distant to its boundaries, i.e., a thinned token. Thinned tokens resulting from the medial-axis transformation are compared with each other and thinned tokens having substantial similar shapes are grouped together. Once the thinned tokens are classified into their respective groups, a representative token of each group is selected for storage along with an assigned token number and the position that each of the corresponding thinned tokens appeared in the work of content. The thinned token-based file may then be transferred to another computing device using less bandwidth than the original token-based file. A page image of the original work can then be reconstructed by the receiving computing device from the thinned token-based file.