A novel approach to detect edge- and line-like events in images was developed on the basis of linear filters that are in quadrature phase. The responses of such filters are ambiguous in that an image event (e.g. an edge or a line) usually produces a multitude of response extrema. The method developed disambiguates this representation by suppressing filter responses that are not directly related to the input feature. Suppression is derived from first directional derivatives of the modulus representation (Euklidian norm of quadrature pair filter outputs). The effect of suppression is that only the filter responses that represent the true position of a feature will remain. The scheme also includes an enhancement component on the basis of second directional derivatives of response modulus. This favors isolated edge or line features over textured areas or featural interferences.