Generally the image may be separated into differential characteristics and details and put the symbol $q_j$ in accordance to each characteristic. Under such approach the vector of the image $q = (q_1, q_2, ..., q_n)$ will correspond to the whole image (figure 1). It is clear that instead of the vector it can be examined his end – the point in the multidimensional space. Therefore, to each image we put a vector or point. In Figure 1 the ends of the vectors indicate the image of the prototype, while the points located in the vicinity of the ends are slightly different from the prototype and still have similarities to him.

As our programme operates with quite primitive images, it is quite sufficient to present the image as a vector, which elements are 0 and 1 (0 -uncolored pixel, 1 – colored pixel ).
Under the same pattern the vector of recognition image is built.
$$ \dot{q} = \sum_{k} \lambda_{k} v_{k} (v_{k}^{\dagger} q) - B \sum_{k' \neq k} (v_{k}^{\dagger} q)^{2} (v_{k}^{\dagger} q) v_{k} - C(q^{\dagger} q) q + F(t) \tag{1} $$