Orientation estimation: Conventional techniques and a new non-differential approach

The estimation approach discussed in this paper is based on a signal-theoretic and statistical analysis of the notion of orientation. In contrast to other approaches, it does not require the computation of gray value gradients, or the power spectrum of the given signal patch, or quadrature filter outputs, but directly estimates a small central part of the autocovariance function (acf) of the signal and derives the sought orientation from the direction of main curvature in the acf origin. The adaptation to the frequency-dependent signal/noise ratio of the signal is performed here by linearly filtering the acf, but not the signal itself. This is in contrast to the usage of pre-filters which are necessary in conventional approaches in order to compensate for the limited capabilities of the discrete differentiation filters. This shortcut which is possible as a consequence of the Wiener-Lee theorem is the key to considerable savings in computational effort.