Pattern Classification with Rejection Using Cellular Automata-Based Filtering

In this article we address the problem of contaminated data in pattern recognition tasks, where apart from native patterns we may have foreign ones that do not belong to any native class. We present a novel approach to image classification with foreign pattern rejection based on cellular automata. The method is based only on native patterns, so no knowledge about characteristics of foreign patterns is required at the stage of model construction. The proposed approach is evaluated in a study of handwritten digits recognition. As foreign patterns we use distorted digits. Experiments show that the proposed model classifies native patterns with a high success rate and rejects foreign patterns as well.