. Why do Sobel and Prewitt kernels combine a difference in one direction with smoothing in the perpendicular direction?
Averaging over rows/columns while differencing cuts down the noise response substantially compared to a bare difference kernel.
. In the Canny edge detector, what is the purpose of non-maximum suppression?
Non-maximum suppression keeps a pixel's gradient magnitude only if it's a local maximum along the gradient direction, producing thin edges.
. What does Canny's hysteresis thresholding do with pixels between the low and high threshold?
Hysteresis links up weak-but-real edge segments that connect to strong edges, while suppressing isolated noise responses.
. In the Hough transform, what does it mean when several edge points' sinusoids in (rho, theta) space all cross at the same point?
Each edge point traces a sinusoid of all lines through it; collinear points' sinusoids intersect at the (rho, theta) of the line they share.