Arrays of low-level inequality based feature detecting cells

We examine how we can combine several simple feature detectors for the recognition of more complicated features. In particular, we develop an algebraic theory which can be used to examine how the templates of feature detectors can be combined into larger templates, or conversely, how complicated templates can be decomposed into simpler ones. The essential part of the theory is the use of Groebner bases to investigate the structure of the templates. As a result, we can state the conditions under which a set of templates can recognize a feature without making errors. In addition, we examine how templates can be added to improve the detector