Introduction of N-gram into a run-length encoding based ASCII art extraction method

Abstract

As ASCII arts can be noise for natural language processing, ASCII art extraction methods can be used to remove them from text. A run-length encoding (RLE) based ASCII art extraction method proposed in our papers uses compression ratio by RLE for recognition of ASCII arts as ASCII arts tend to be compressed small by RLE and non-ASCII arts do not. It is because same characters tend to occur successively in ASCII arts but they do not in non-ASCII arts. Small ASCII arts, however, are not compressed as small as large ASCII arts. In this paper, we add the occurrence number of n-gram of ASCII arts in text into the RLE-based method as a new text attribute to cope with small ASCII arts. Our experimental results show that the new attribute improves the F-measure but it adds language-dependency into the RLE-based method though it is desirable that ASCII art extraction methods are language- independent.