Abstract: The present invention relates to a method of reducing the quantity of overhead data in image coding wherein the image is subdivided into partial image regions and these partial image regions are characterized by classes. To determine the class, the partial image region is transformed by discrete transformation into the spectral range and the absolute values of the spectral coefficients are then processed in a manner such that all absolute values are compared with their visual discernibility thresholds, singular coefficients are neglected and the result can be represented by term variables, i.e. by a minimum quantity of different individual coefficients. The classes derived from a counter and from a class decoding gate array are compared with the pattern representing the term variables until the first congruent class is found. The associated class code is forwarded for further processing.