Lattice masking and auto-association for recalling color images in the presence of noise

Lattice associative memories are artificial neural networks for which the storage and recall stages, given a finite set X of exemplar images, are defined with lattice algebra operations. Two dual canonical auto-associative memories have been introduced, the min-memory W_xx and the max-memory M_xx, capable to recall approximations to exemplars from corrupted inputs. It turns out that the min-memory is robust to erosive noise and the max-memory is robust to dilative noise; however, neither one of these memories is able to cope with images degraded by random noise represented as a mixture of erosive and dilative noise. A hybrid procedure based on noise masking and two measures is developed here to endow lattice auto-associative memories with color image recall capability for inputs distorted by additive random noise.