ihoecken wrote:

May be I'm wrong, but all Fuzzy-Logic algorithms I know are randomized.

No, this is incorrect. The "fuzzy" in fuzzy logic refers to graded truth values, not randomness or chance. This is an important point.

ihoecken wrote:

Take a look at measurement of polygons. A "coin" is dropped to an rectangular area. If it falls on the polygon it's counted. Afterwards you can calculate the size dividing the counted coints by all coins thrown. Knowing the size of the rectangular, you can calculate the size of the polygon. This is a very simple fuzzy logic algorithm ...

No, this is called a "Monte Carlo" algorithm, and is not fuzzy at all.

ihoecken wrote:

This is a very simple fuzzy logic algorithm, but all others I know (for picture stabilization for cameras) are working in the same way. There is always a probabilistic algorithm in the background and so it's non-deterministic. The definition of fuzzy logic has nothing to do with probabilistic working (it's about multistate bits), I know, but I don't know algorithms which uses multistate bits without probabilistics.

Although systems which employ fuzzy logic might employ randomness somewhere, the "fuzziness" of fuzzy logic is not probabilistic. I refer interested parties to: Usenet Fuzzy Logic and Fuzzy Expert Systems FAQ The Fuzzy Systems Handbook by Earl Cox (ISBN-13: 978-0121942700) Fuzzy Sets and Fuzzy Logic by George J. Klir and Bo Yuan (ISBN-13: 978-0131011717)

Assuming that these are simple geometric shapes on relatively simple backgrounds (as opposed to real objects against cluttered backgrounds), then I'd suggest: 1. Determine which pixels are foreground (shapes) and background (everything else) 2. Identify individual blobs (connected sets of foreground pixels of the same type) 3. For each blob, extract meaningful features 4. Train a classifier, based on the extracted features 5. Test the system on new images 6. Celebrate! In step 2, try using a flood fill. In step 3, there are many features one might try, such as perimeter-to-area ratio. In step 4, the classifier could be any learning algorithm: neural network, linear discriminant, etc. -Will Dwinnell Data Mining in MATLAB

The basic steps will be something like: 1. Locate and extract the license plate within the image (not always easy, especially if the vehicle is moving) 2. Clean and enhance the license plate image (de-skew, de-noise, improve contrast, etc.) 3. Isolate individual glyphs (characters) 4. Extract informative features from characters 5. Recognize individual characters Step 5 could be performed any number of ways: neural network, discriminant analysis, k-nearest neighbors, etc. -Will Dwinnell Data Mining in MATLAB