# Date and Time

The exercises take place on

Monday, 16:15-17:00

Wednesday, 16:00-16:45

in room 09.150 (computer science building, seminar room of the Pattern Recognition Lab).

# Exercise sheets

Please prepare the exercises at home. We will put the exercises online a couple of days (typically more than 5) prior to the class.

 Exercise 1 Statistics Exercise 2 Quantization Exercise 3 ML-Estimation and Binarization (you will also need the test image "object"!) Exercise 4 Image enhancement (and test image "moon") Exercise 5 Fourier series, histogram equalization (featuring Lena in color and black and white \o/) Exercise 6 Everything about moments! (with the test images momented and unmomented) Exercise 7 Linear predictive coding Exercise 8 Wavelet decomposition - Note that this is a coding-only exercise! Q: Programming only? Should I ignore it? A: Omg, NO!  --> better go ahead and download the supplemental material :) Exercise 9 Principal component analysis and testing data Exercise 10 Branch&bound (note that the fixed point theorem has been removed) Exercise 11 Classification (-> re-review this work sheet, Exercise 28 has been clarified & expanded!)

# Supplemental Material

Some facts on the Fourier transform.

Some facts on statistics.

A brief explanation on the variance of the estimated mean (Exercise 9). A participant had a question on an identity that can be derived using the Bienayme formula. So relax - everything is sane :)

...and a comment on the repetition of the first row at the end of the matrix in the discrete Haar transform (worksheet 8, wavelet transform). The result of conv2 should contain as much entries as the input (in order not to lose information during the transformation). Thus, as we have a filter of width 2, we would lose one line in the output, if the wrap-around of the matrix entries weren't added. However, in our example application, this loss "would not hurt much" (-> try it), but it is still imprecise.