A method of image recognition based on the fusion of reduced invariant representations: mathematical substantiation

An approach is proposed to a solution of image recognition problems based on the following principal issues: a) images are represented by multiple partial models widely used in pattern recognition - feature sets; b) algorithms are multiple classifiers: each algorithm uses its own data - some partial image model; c) there are two kinds of fusion - data (partial models) fusion and algorithm fusion; d) fusion processes of both kinds are implemented by algebraic techniques in the framework of descriptive theory of image analysis. The specific aspects of the proposed approach: a) a partial image model (a feature set) includes only image invariants; b) the concept of image equivalence is a base for recognition through exploiting the partial image models; and c) the final solution is obtained by the algorithm fusion methods using algebra of images and algebra of algorithms.