A neural network algorithm for complex binary image recognition requiring no presegmentation

Conclusion

The suggested mechanism of competition between categorical neurons has the following advantages over alternative mechanisms:

• --it resolves complex images into composing reference images without learning all combinations of the reference images;

• --it is consistent with the neurophysiological findings of a contribution of intercalary neurons to inhibitory feedback chains;

• --it requires less RAM volume to store feedback weighting factors, this being particularly important for computer modeling.

We would like to thank Yu. N. Zolotukhin for patient and consistent support of this research.