Score: 0 / 4
. What does the Fourier transform (Lesson 15) fail to tell you, that wavelets and Gabor filters are designed to capture?
Explanation: Sine wave basis functions extend across the whole image, so the Fourier transform loses spatial localization; wavelets and Gabor filters use spatially localized oscillations instead.
. In the Haar wavelet transform, what do the approximation (a_k) and detail (d_k) coefficients represent?
Explanation: a_k is the (scaled) sum of a sample pair, i.e. a local average; d_k is the (scaled) difference, i.e. a local detail.
. In a one-level 2D wavelet decomposition of an image, what does the LH subband respond to?
Explanation: LH is low-pass along one axis and high-pass along the other, which makes it sensitive to edges in one direction -- unlike HH (diagonal/corner detail, high-pass both ways) or LL (the blurry low-res approximation, low-pass both ways). The exact axis-to-direction mapping (LH = horizontal or vertical edges) depends on the row/column convention used.
. What does a Gabor filter's response tell you when the filter is tuned to one orientation and applied to bars drawn at several different orientations?
Explanation: The response matrix in the lesson is diagonal-dominant: each filter's peak response lands on the bar matching its own tuned orientation, demonstrating orientation selectivity.