Pattern detection in spectrograms by means of Cellular Neural Networks

The presented paper introduces a method that allows for detecting complex patterns in spectrograms by means of a CNN. The method is based on two object recognition ideas adopted from the deformable grid paradigm: a distributed representation of object prototypes and a prototype-image matching for establishing a class membership. A proposed CNN pattern detection algorithm exploits specific properties of spectrogram images by differentiating rules for processing of temporal and frequency information. Isolated utterance detection has been used as a sample application for evaluating the proposed algorithm performance