Processing spectrograms as digital pictures to separate competing line signatures

Abstract The problem of extracting individual signatures from a (digital) spectrogram containing two or more line signatures amidst random noise is discussed. It is assumed that each signature family contains a dominant line and the method proposed is based on using line enhancement and grouping procedures to isolate these dominant components. Using the extracted lines, a family of two-dimensional matched filters are designed for each signature and used to detect the weaker members of the family. Cleaned-up low-entropy spectrograms of each source can then be produced.