338
1 |
SETALSEy —ALEOELENDS om Techniques cce Hardware
| ) | i
Fig. 5: Interrelations of components
An outline is given in the following sections of the input data,
matching strategies, algorithms, techniques, hardware, and the qua-
lity indicators.
1l. Input data
Matching is applied to pre-processed image data, i.e., original
and/or synthesised (e.g., envelop, compressed second differences,
etc.). Target segments can be laid-out in single or several pat-
terns, which can be regular, semi-regular or irregular. Moreover,
the lay-out can be in single or multiple hierarchical levels (2). A
segment can contain full image data or only the compressed (rele-
vant) data. Individual pixel intensity values can be unweighted or
weighted, e.g., according to information content and/or according to
pixel location in the target segment. ;
Another differentiation has been made between non-distinct image
segments and distinct image features (primitives).
2. Matching stategy
The strategy refers to matching itself and to self-adaptive capabi-
lity. Matching procedures can be single (homogeneous) or multiple.
The latter can be sequential (i.e. proceeding from distinct points
to edges and further to areas) or parallel (e.g., of several match-
ing trials, several full matches or several groups of matches).
Sequential and parallel calculations can be combined.
Adaptive operations use feedback or feedforeward control or a combi-
nation of both. They usually imply data resampling, which can be
relative or absolute, and may or may not involve the terrain model
geometry (4).
3. Matching algorithms
In accordance with the strategies, a distinction can be made between
single and multiple matching algorithms. These may or may not permit
inclusion of the external information (vide III.11).
The algorithms can be statistical, deterministic or mixed. The com-
monly used are statistical algorithms, e.g., correlation, covarian-
ce, etc. Deterministic algorithms can be applied to extracted conju-
gate image features (or signatures), or to calculation of specific
image parameters. A mixed algorithm is, for example, an affine
transformation by least squares adjustment, of a target into the
corresponding conjugate segment (5).
4. Matching techniques
The techniques are governed by the strategy, algorithms, and the
available hardware. Commonly applied is the similarity assessment of
conjugate segments by sequential search or by image (intensity)
transformation into another domain (vide chapter IV.5).
339