Neural networks for image analysis
Chris M. W. Daft · IEEE Symposium on Ultrasonics · 2002
The author discusses three applications of neural networks to image processing: the silicon retina of C.A. Mead and M.A. Mahowald (1988); a method of echo inversion described by N.H. Farhat and B. Bai (1989); and the fingerprint recognition method of M.-T. Leung et al. (1990). Mead's work is significant because it forms a front-end for image analysis with a morphology modeled on the eye. Farhat's group has down work on the use of neural networks in image reconstruction. A description of some of this is included because it concerns the regularization of an inverse problem, which is similar to many ultrasonic, medical, and geophysical imaging applications. The method of M.-T. Leung et al. is noteworthy not only as an example of neuromorphic image processing, but also because the representation of the data is inspired by measurements of the mammalian brain. Current areas of research in ultrasonic applications of neural networks are discussed.>