General waveform shape analyser
Paul Lister, Matthew Bishop · IEE Proceedings E Computers and Digital Techniques · 1988
The paper describes a general waveform shape analyser (Wavenet), and a system that uses the representation that it forms for a particular waveform application (Synnet). Wavenet incorporates ideas from models of low-level vision, including Gestalt grouping and the search for symmetry and structure. The novel design includes a lattice network of nodes combined with an attributed grammar, which hunts for perceptually salient structures in overlapping sections of the waveform. The parallel multiscale representation formed by Wavenet is used by the second system, Synnet, to find instances of clinically significant EEG shapes. Synnet chooses the best of the wavenet primitives, using relaxation, and combines them syntactically into higher-level shapes, assigning confidence values calculated from their attributes. Wavenet and Synnet were able to find EEG spikes and waves, of various sizes and with superimposed noise, in a variety of EEG data.