Neural cell classification by wavelets and multiscale curvature (Q1271517)

A new approach to automatic classification of ganglion cells using wavelet transforms, curvature and standard pattern recognition techniques is proposed. The wavelet transform of a characterization of the outer contour of the cells is calculated and the wavelet energy or the bending energy are used to define a set of measures used to discriminate among shape features. The advantages of the two energy techniques are analysed. The proposed technique can be applied to other classes of biological shapes, e.g. mammograms.