Statistical Features for Text-Independent Writer Identification