Anmerkungen:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Beschreibung:
The increasing use of high-resolution image sensors in both stationary and mobile applications require improved image recognition algorithms. The Hausdorff distance is a measure of the likeness of two sets of points, and can be used to determine the resemblance of two sets of image points. However, is not widely used. Therefore, this dissertation deals with a method of using the Hausdorff distance to determine the resemblance of image regions. We introduce a suitable model to describe linear deviations. We show how to compensate for these linear deviations and use a probability distribution for their classification. We give bounds for the non-linear deviations and minimize noise. Our starting point is the mathematical description of the mentioned criterion for deviation. We calculate deviation of pairs of image points and encode it in a three-dimensional vector field. This vector field also contains the directions in which differences are decreasing. Using this information we obtain a probability density function which gives a measure of similarity. We interpret transformation of distances as a stochastic vector process. This opens up new directions for compensating for geometric displacements of image regions. We then use our deviation model in a control loop to minimize linear deformations. Our filter proves robust with respect to Gaussian noise. We show equivalence of the metrics $d_2$ and $d_1$ for Gaussian noise. This is the main prerequisite for hardwired speed improvements, and we use it in the design of a distance processor. With the help of our distance processor we show that our control loop is stable. When using the directional information in the distance vector field we observe an increase of the correlation of image regions in question. A correction transformation greatly reduces sensitivity to noise of the Hausdorff distance. Our resource-friendly VHDL design allows the real-time calculation of distance vector fields with current FPGAs. The stability of our control loop improves when we include ...