Milestones
| 1 | 2 | 3 |

3. Implementation of String Alignment methods to match boundary information

The information strings that are the result of the computations in the second step are now used to compute an alignment score that gives a hint on the silmilarity of images. For this purpose several alignment algorithms have to be implemented.


a) Alignment algorithms

SA_extract returns the boundaries and strings for a given image. To get a similarity score for two of these strings the following methods are implemented: SA_globalAlignment, SA_localAlignment, SA_repeatedMatchesAlignment and SA_overlapMatchesAlignment.

These functions compute a similarity score for a possible alignment using different methods. These four methods are subject to further testing to see which one gives the best result.




b) Visualisation of alignments

The mentioned functions also return an encoded alignment that allows to reconstruct the optimal alignment. This is then used in the function SA_visualizeAlignment to show the alignment in an image together with the boundaries. The following picture shows an example of such a visualisation:



The red and blue lines represent the boundaries from which the information strings where extracted. The dots show the positions at which the polarscales where used. The green lines show which points were aligned.
SA_visualizeAlignment can now be used to adjust the parameter of the alignment algorithms and to see which algorithm leads to the best results.




c) Alignment test function

To make the procedure of adjusting the alignment parameter easier the function SA_alignmentTest can be used.




d) Creation of an alignment score matrix

The final result of the alignments should be a matrix of scores in wich all elements of two lists of strings (from a list of test images) are aligned to each other. This matrix is returned by the function SA_alignmentMatrix.




Copyright © 2007 Thomas Oskam, ETH Zürich. All rights reserved.