Structural and Syntactic Methods in Line Drawing Analysis: To Which Extent Do They Work?

Syntactic methods in pattern recognition have been around for more than 20 years now [22, 23], and numerous applications have been done with these basic techniques. Structural pattern recognition is an even vaster field [47] (syntactic methods can actually be considered to belong to it), where many methods exist and have been used for various practical applications. These methods have been extensively used in the application domain of character recognition, document analysis and graphics recognition, as appears from our review in § 2. It may therefore seem strange to ask today if these methods really work. But it has been our experience that when designing larger graphics recognition systems, we come to a number of what can be called “border problems”, which at first glance appear to be suitable precisely for this family of recognition methods, but where expected and more or less unexpected hindrances make life difficult for the recognition methodology. We present some of these problems in § 3. As people use more or less strict definitions of what structural and syntactic pattern recognition is, let us first introduce our own understanding of these concepts. Structural methods are based on the relational organization of low-level features into higher-level structures. These structures are then matched with model structures. These methods are usually associated with graph matching techniques and structural organization of simple features into perceptual groups, based on a number of relationships between the features. We summarize the main idea by citing Suetens et al. [54]: