Abstract
Analysis of sentences in a natural language is often based on similar methods as analysis of formal languages. Analogically, analysis of pictures could be based on analysis of formal picture languages.
However, the field of formal picture languages is not developed enough for this purpose. This paper presents several models of automata accepting two-dimensional languages and outlines their learning capabilities.
Further, it examines the possibility of transforming a two-dimensional language into a one-dimensional language and applying machine learning techniques in a single dimension. In this paper, we propose a new representation for formal picture languages consisting of two components - a picture-to-string function and a string language.
The function rewrites any two-dimensional picture into a string. A picture language is then the set of all pictures that the function maps into the given string language.
Using this representation, picture languages can be learned by applying methods of grammatical inference for string languages.