The procedure of selection of the objectlike areas from the images of the basis of geometrical features is considered in this paper. The local anisotropic features of the images are used as the geometrical features. The problem of selection of the objectlike areas is considered here as the problem of extraction of image areas with the properties are close to the properties of sample subimage (object), i.e. it is the subdivision of the image into `background' and `object' parts. General sense of the approach presented here is transformation of image to the such kind, when the background and objectlike areas of the image are maximally divided, and determination of rule for division of the areas. The localglobal strategy of the image analysis, so- called effect of a rebound, enables to determine parameters of the description of the image, which are optimum from this point of view. The description of the areas, we are interested in, is defined by the appropriate samples, the origin of which is determined by the given problem. Real, graphic and synthesized images can be used as samples. The given technique is based on properties of image only, and it enables to select the areas with a given configuration of the standard objects. Self-organizing noninstructable procedure of selection can be considered as a stage of object recognition.