Investigation of reliability of combinatorial-metric algorithm for recognition of $n$-dimensional group point object in hierarchy features space

The scientific research of reliability of combinatorial-metric algorithm for multi-dimensional group point objects recognition in hierarchically organized features space is considered in the paper. The nature of reliability indicator change is examined, as an example, using multilevel descriptions of simulated and real objects under the condition that recognition results obtained at one hierarchy level are used as input data at  next level.
A priori uncertainty of a view angle, composition incompleteness and coordinate noise of objects determine the combinatorial procedures of quantifiable estimation of proximity of multidimensional GPO, presenting the object of recognition to a particular class.
The stability of the recognition algorithm is achieved by the possibility of changing  strategy of making a classification decision. For this purpose, we use the representation of a group point object at the lowest level of the hierarchy in the form of: sample, composition of sample elements or a complex a priori indicator. In order to increase the recognition accuracy, it was proposed to use the search of recognition results at  low levels of the hierarchy. The experimental dependences of a priori and a posteriori reliability indicators for various conditions for measurements and states of recognition objects are provided in the paper.