Real images contain symmetric Gestalten with high probability. I.e. certain parts can be mapped on other certain parts by the usual Gestalt laws and are repeated there with high similarity. Moreover, such mapping comes in nested hierarchies - e.g. a reflection Gestalt that is made of repetition friezes, whose parts are again reflection symmetric compositions. This can be explicitly modelled by continuous assessment functions. Hard decisions on whether or not a law is fulfilled are avoided. Starting from primitive objects extracted from the input image successively aggregates are constructed: reflection pairs, rows, etc., forming a part-of-hierarchy and rising in scale. The work in this paper starts from super-pixel primitives, and the grouping ends when the Gestalten almost fill the whole image. Occasionally the results may not be in accordance with human perception. The parameters have not been adjusted specifically for the data at hand. Previous work only used the compulsory attributes location, scale, orientation and assessment for each object. A way to improve the recognition performance is utilizing additional features such as colors or eccentricity. Thus the recognition rates are a little better.
