Building Digital Libraries from Paper Documents, Using ART Based Neuro-fuzzy Systems

摘要

In this paper a new neuro-fuzzy system is proposed for both tasks of document analysis and Optical Character Recognition. FasART (Fuzzy adaptive system ART based) inherits the stability, flexibility and modularity properties of ART supervised models, but with a formal description as a Fuzzy Logic System, and increased functionality. On the other hand Recursive FasART permits us to exploit context information, crucial aspect in document understanding. Satisfactory experimental results are presented for the global application of building a digital library of scientific papers, giving special emphasis on the creation of links between items in table of contents and paper first pages.