Invariant pattern recognition against defocus based on subband decomposition of the filter

The discrimination capability (DC) is an important quality parameter in pattern recognition. If the input scene suffers some degradation, this parameter decreases. A source of degradation could be the imaging system, in particular the obtained image could be out of focus. In this paper we analyze single and multichannel methods to obtain a DC invariant to the defocus. They are based on the Laplacian pyramidal decomposition of the filter. In the case of multichannel procedures we analyze several nonlinear methods to combine the information given by each channel.