Adaptive neural network for pattern recognition of 2-D image under affine transformation

Presents a method for recognition of a 2-D image under affine transformation, which can be used as the preprocessing unit in an adaptive neural network. The affine transformation is decomposed into six basic transformations: x-direction and y-direction translations, rotation, x-direction and y-direction expansions (or compressions), and x-direction shear. In order to recognize a 2-D image under affine transformation, the authors introduce the normalized form of an image and design the normalizer by which one can transform an arbitrary input image to its normalized form. The recognition of the images can be done by comparing their normalized forms with a neural network identifier. Some experimental results are reported