(b) Keypoints with a high
Tr(H)^2/Det(H)
ratio are rejected.
Justification:
A high
ratio indicates an edge response, and since edges are not as stable as corners for
keypoints, points with a high
Tr(H)^2/Det(H)
ratio are rejected.
(d) Keypoints are rejected if the determinant of the Hessian matrix is negative.
Justification:
A negative determinant of the Hessian matrix indicates a saddle point
rather than a local maximum or minimum, which is not suitable for a keypoint.
Saddle points are not extremal points and therefore are rejected.
Orientation Assignment in SIFT:
What is the goal of orientation assignment for keypoints in the SIFT algorithm?
(a) To ensure the descriptor is robust to changes in lighting.
(b) To make the descriptor invariant to image rotation.
(c) To increase the contrast of the descriptor.
(d) To assign a unique identifier to each keypoint.
Justify your answer.
(b) To make the descriptor invariant to image rotation.
Justification:
The orientation
assignment in SIFT assigns a consistent orientation to each keypoint based on the
gradients around it. This process ensures that the keypoint descriptor is rotation-
invariant, which means the descriptor will not change even if the image is rotated.
Histogram Creation for Descriptors:
When generating a SIFT descriptor, how is the histogram of orientations created?
(a) By taking a 16x16 pixel area around a keypoint and computing 36 bins for each
pixel.
(b) Through a 4x4 pixel block around a keypoint, with 8 orientation bins.
(c) Using a circular window around the keypoint, normalized by a Gaussian.
(d) By computing one bin per degree in a 360° range around the keypoint.
Justify your answer.
(b) Through a 4x4 pixel block around a keypoint, with 8 orientation bins.
Justification:
A SIFT descriptor is created by first dividing the region around a keypoint into 4x4
blocks and then creating a histogram of gradient orientations for each block.
Typically, 8 orientation bins are used for these histograms.
Descriptor Length in SIFT:
How many numbers are obtained from the 16x16 pixels around a keypoint to form a
SIFT descriptor?
(a) 16
(b) 64
(c) 128
(d) 256
