AUTOMATIC multilevel form colour dropout using graph cuts

We present a new efficient solution to the problem of form colour dropout, required for automated processing of machine-printed forms. Unlike previous methods, our solution does not require any prior information about the hardware, scanner settings, colour space parameters, document template information - indeed we require no prior information at all. Our approach demonstrates a new application of graph cuts; we apply a simple graph cut on the scanned form image in order to drop out fields, which have been printed in pre-defined colours unknown to our system. We effectively separate the machine-printed and hand-printed data from the template, thereby making simple the remaining steps involved in processing the document.