Comparison of two ASCII art extraction methods: A run-length encoding based method and a byte pattern based method

Text based pictures called ASCII art are often used in Web pages, email text and so on. They enrich expression in text data, but they can be noise for natural language processing and large ASCII arts are deformed in small display devices. We can ignore ASCII arts in text data or replace them with other strings by ASCII art extraction methods, which detect areas of ASCII arts in a given text data. Our research group and another research group independently proposed two different ASCII art extraction methods, which are a run-length encoding based method and a byte pattern based method respectively. Both of the methods use text classifiers constructed by machine learning algorithms, but they use different attributes of text. In this paper, we compare the two methods by ASCII art extraction experiments where training text and testing text are in English and Japanese. Our experimental results show that the two methods are competitive if training text and testing text are in a same set of languages, but the run-length encoding based method works better than the byte pattern based method if training text and testing text are in different sets of languages.