Activity Driven Update in the Neural Abstraction Pyramid

Abstract

The Neural Abstraction Pyramid is a hierarchical neural architecture for image interpretation based on image pyramids and cellular neural net- works and inspired by the principles of information processing found in the visual cortex. In this paper we extend the model by describing a parallel mechanism of bottom-up attention control, the Activity Driven Update of processing elements. We apply this mechanism to the binariza- tion of handwritten ZIP-codes in a real-world application. The experi- mental results indicate that updating only a fraction of the processing elements is sufficient for good binarization. Both speed and performance of the application were improved with the new method.