About some properties of estimation of method of character recognition, based on polynomial regression

Shown that at recognition of handprinted figures on base, coinciding with base of education, average estimation of recognition for each of considered symbols decreases monotonously on measure 'removing' from ‘average’ polynomial vector, and for ‘1’ first monotonously decrease, and then monotonously increase and take maximum rating 255. Level of noises in these dependences greatly lower than in dependences at deviation from 'average' raster. In result numerical research it was possible to show that base of single symbols, consist of two subbase, each of which has own 'average' vector and (raster). Researched recognition method, trained on sets of all images ‘1’, expose estimations of recognition, behaviour of which at removing from each of these two 'average' vectors are not distinguished from having place for bases of other figures with one ‘average’ vector and (raster).