A syntactic method for object recognition from a color image

This paper presents a method for object recognition from a color image by using syntactic pattern analysis. From a digital image, the contours of the input objects are segmented and encoded into a string of Freeman codes. A polygonal approximation and a circular arc approximation are used to give a significant and concise shape representation of the analyzed objects. To recognize the shape of the objects, the authors propose a new method based on the idea of the attributed grammar. By using this new method, the strings of primitives representing an input object can be matched against strings of primitives representing each model or reference object. Finally, the input object is classified in the same class as the model that is the “best” match to the input. Any existing model can be generalized from a set of input objects as examples, which have the same characteristics in some sense