Recognition of characters subject to spatially independent transformations

Abstract Binary valued characters, defined over a discrete subset of two-dimensional space, are assumed to be contaminated by spatially independent random errors at each point in space. The errors are described as random transformations. That is, a probability distribution over the possible transformations is assumed at each point. With this definition of the contaminating error process, the optimal Bayes recognition procedure is developed. The optimal recognition rule is linear in the character observations.