Pattern recognition based on morphological shape analysis and neural networks

Morphological transformations are efficient methods for shape analysis and representation. In this paper two morphological shape descriptors are described for object feature representation. Neural networks are then employed for object recognition and classification. Various coding schemes and training procedures are examined in order to achieve a high classification performance. Complete classification and recognition schemes are proposed, which are shown to work satisfactorily even for objects distorted by quantization noise as well as partially invisible objects. Classification results are compared with those obtained by other shape descriptors which use conventional recognition and classification schemes.