That was not what I wanted.
if I have in text '�' and é I want to obtain both of this. With your methods I get only �.
Here are two solutions for this problem that I found since I post this.
1. in characters(char ch[], int start, int length) the ch[] contains all the text of the input *except* when a character reference is procesed and then ch[] = {'�'} (this � is comming from a é and not from �)
Found this from looking into the xerces's sources.
2. after talking with somebody at apache I found that there is a not documented feature:
http://apache.org/xml/features/scanner/notify-char-refs. Put this to true and you will get character references reported in the same way as entities.