So, it's more complicated to work on the xml schema ? if so, what better could I do ?

>>>>>>>> 
If your schema is generated from a DTD, then on reflection, I think it might be simpler to analyse the schema at source XML level. The converter will probably generate a very stereotyped schema with a regular structure, which you can exploit when reading it. You won't have to deal with problems such as element wildcards, named model groups, local element declarations, substitution groups and so on, because they won't exist. The full schema model has a lot of complexity to handle such things.
 
Michael Kay
http://www.saxonica.com/