Page - 175 - in Document Image Processing

J. Imaging 2017,3, 62 Figure2. Syntheticdocument imagegeneration. (Left) originaldocument image. (Right) synthetic document imagegeneratedautomaticallywith therandomtext“Loremipsum”. Theautomatically generated image lookssimilar to theoriginalone. Theresult is stillperfectible.Here,as theoriginal imagesuffers fromlocaldeformations, the characters extracted tobuild the fontarequitedifferent andmay look too randomwhenassembledon the syntheticdocument. Abetter font extractionor compositionusingthecontext tochoosenewcharactersmayalleviate thisproblem. 3.DocumentDegradationModels Physicaldegradationduetoageing, storageconditionsorpoorqualityofprintingmaterialsmay bepresentondocuments. DocCreatorcurrentlyproposessevendegradationmodels. Asdetailed inFigure1 (rightpart), all thesemodelscanbeappliedonreal images toextendany document imagedatabase. TheusercaninteractwithDocCreator inorder toset thequantityofdefects togenerate. In the following sections, we describe the main ideas of these seven degradation models. AsDocCreator is anopensource software, readers canconsult the source code togetmoredetails about the implementationof thesemodels. 3.1. InkDegradation DocCreatorprovidesagrayscale inkdegradationmodel (detailed in [41]) able to simulate the most commoncharacterdegradationsdue to the ageof thedocument itself andprinting/writing process, such as ink splotches,white specks or streaks. Thismodel locallydegrades the image in theneighbourhoodof the charactersboundaries. Noise is thengenerated to create somesmall ink spotsnearcharactersor toerasesomecharacters inkarea.Contrary to thewellknownKanungonoise model [47] thatworksonlyonblackandwhite images, thisdegradationmethodcanprocessgrayscale images. SeeFigure3 foran inkdegradationexample. 175