Bookmark and Share

Computer Vision Metrics: Chapter Four (Part D)

Register or sign in to access the Embedded Vision Academy's free technical training content.

The training materials provided by the Embedded Vision Academy are offered free of charge to everyone. All we ask in return is that you register, and tell us a little about yourself so that we can understand a bit about our audience. As detailed in our Privacy Policy, we will not share your registration information, nor contact you, except with your consent.

Registration is free and takes less than one minute. Click here to register, and get full access to the Embedded Vision Academy's unique technical training content.

If you've already registered, click here to sign in.

See a sample of this page's content below:

For Part C of Chapter Four, please click here.

Bibliography references are set off with brackets, i.e. "[XXX]". For the corresponding bibliography entries, please click here.

Brisk Patterns

The BRISK descriptor [131] point-pair sampling shape is symmetric and circular, composed of 60 total points arranged in four concentric rings, as shown in Figure 4-10. Surrounding each of the 60 points is a sampling region shown in blue, the sampling regions increase in size with distance from the center, and also proportional to the distance between sample points. Within the sampling regions, Gaussian smoothing is applied to the pixels and a local gradient is calculated over the smoothed region.

Figure 4-10. (Left) BRISK concentric sampling grid pattern. (Center) Short segment pairs. (Right) Long distance pairs. Note that the size of the region (left image) for each selected point increases in diamter with distance from the center, and the binary comparison is computed from the center point of each Gaussian-sampled circular region, rather than from each solitary center point. (Center and right images used by permission © Josh Gleason[143])

Like other local binary descriptors, BRISK compares pairs of points to form the descriptor. The point- pairs are specified in two groups: (1) long segments, which are used together with the region gradients to determine angle and direction of the descriptor, the angle is used to rotate the descriptor area, and then the pair–wise sampling pattern is applied;(2) short segments, which can be pair-wise compared and composed into the 512-bit binary descriptor vector.

ORB and BRIEF Patterns

ORB [134] is based in part on the BRIEF descriptor [132,133], thus the name Oriented Brief, since ORB adds orientation to the BRIEF method and provides other improvements as well. For example, ORB also improves the interest point method by qualifying FAST corners using Harris corner methods, and improves corner orientation using Rosin’s method [61] in order to steer the BRIEF descriptor to improve rotational invariance (BRIEF is known to be sensitive to rotation).

ORB also provides a very good point-pair training method, which is an improvement over BRIEF. In BRIEF, as shown in Figure 4-11, the sample points are specified in a random distribution pattern based on a Gaussian distribution about the center point...