Image labeling using hopfield neural NE

الملخص EN

Tlfis paper considers ،the application of neural networks to the problem ol image labeling.

It is assumed that an image has been segmented into regions and that predefined uniary properties as well as binaty relationships between these regions have been extracted from the segmented image.

It is also assumed that the knowledge base of object models contains corresponding unary features of object models.

A Hopfield optimization network is used as a matching mechanism based on an energy function that compute the hamming distance between tow neurons (nodes) based on the consistency between object models and image regions with regard to the unary and binary constraints.

A node represents an association between an image region with one of the possible labels.

The results presented in this paper indicate validity of the neural approach and the corresponding energy function used.