Poliakova T. Image classification based on feature spatial structures

Dissertation is devoted to research and development of methods for image classification based on descriptions in the form of a set of characteristic features in times of geometric transformations and spatial interference. Synthesized by the method of comparing the descriptions of images based on modeling the feature invariant structures using the principle of voting. Was developed a model for the description of the construction of objects through the use of spatial structures, taking into account the context and the spatial properties of the elements of description and the subsequent contraction of the resulting set of structures. Further developed a hybrid model of image classification using a comprehensive analysis of different types of structural features in the form of a set of values of descriptors and invariant attributes that are based on their coordinates. Was improved a model of computing similarities of descriptions using the principle of the optimal group matches of their elements by setting the maximum matching between the elements of description that is different from the known models of the principle of establishing a one-to-one correspondences features and provides a minimum criterion of the total distance between the sets of elements of the description of the object and a reference.