Automatic recognition of biological shapes using the Hotelling transform

We present two contour-based techniques, for computerized object recognition that avoid the difficulties due to translations, rotations and scaling. Our techniques do not require any shape representation. The first technique uses the scale-space filtered coordinate functions of contours and their "largest diameters". The second, uses the Hotelling transform of the vector representations of the points of contours. We applied both techniques to recognize human corneal endothelial cells, embedded in a sample of tissue. The results obtained using the Hotelling transform represent a considerable improvement when compared to those previously obtained using the coordinate functions, the curvature function and Fourier descriptors as representations of shape.