Research of vertebrate and invertebrate vision systems has revealed them to be remarkable assemblies of simple cells performing collectively various image processing and analysis tasks. Among these are counted edge enhancement, noise suppression, dynamic range compression, and motion and object orientation detection. These functions are achieved due to the massively parallel structure of these systems and appropriate non-linear inter-cell interactions, among them lateral inhibition. The high degree of connectivity existent in the vertebrate retina is currently beyond reach of integrated implementations; however, even its approximations applied to focal plane arrays can result in enhanced and more sophisticated performance. These approximations are discussed mathematically by means of methods developed for analysis of neural networks. A photoreceptor lateral interaction network, Grossberg's shunting neural network, and a novel modified version of the latter are compared in their effect on spatial nonuniformity noise and edge enhancement. These two qualities are of special interest in the case of infrared imaging. The modified shunting network combines an adaptive lateral signal spread amongst photodetectors with non-linear, multiplicative lateral inhibition. The first effect serves to reduce the effects of spatial noise, while the second, by its differentiating nature, removes low spatial frequencies and enhances high spatial frequency components inherent to the image.