Some experimental results on sparsely connected autoassociative morphological memories for the reconstruction of color images corrupted by either impulsive or Gaussian noise

Sparsely connected autoassociative morphological memories (SCAMMs) are single layer feedforward neural networks that compute either the maximum or the minimum of a finite subset of the input signals. These memories are computationally cheaper than traditional models and have a wide range of applications because they rely only on a complete lattice structure, which is obtained by imposing some ordering on a set. In particular, SCAMMs can be used for the storage and recall of color images. However, there exist several mathematical representations of color images, including the RGB, HSL, and CIELab color systems. Furthermore, the colors can be ordered in many different ways in each system. In view of these remarks, this paper aims at providing some experimental results on the performance of SCAMMs, defined on different ordered color models, for the reconstruction of color images corrupted by either Gaussian or impulsive noise.