Composite document analysis by means of typographic characteristics

Abstract

We have just presented a new method, of regrouping letters and words in homogeneous font families which doesn't necessitate to explicitly recognise the font. This analysis, achieved with the application of one pattern redundancy technique, allows us to extract a part of the logical information which is carried by words typographic features. After having differentiated, grouped together and compared the typographic families, we'll know: - the cardinality of each family, - its grease, slope and size compared to the others families. The study of the typographic families organisation, and of their relative characteristics, will allows us to classify families according to their logical significance, and so to voice, when it will be possible, hypothesis concerning the logical signification of the families. A comparison between the constructed families and the learned grammar, will come to validate or correct the hypothesis, and to label families for which no hypothesis has been voiced. The significance of the method, we have developed, is that each process only depend on the image ; it isn't depend on the document type or on fonts data basis. So this method can be applied to every document type, specially complex and typographically rich documents. An other significance is that our text markers will be use for describing our document in HTML language