Bookmark and Share

Computer Vision Metrics: Chapter Four

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:

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

Local Feature Design Concepts, Classification, and Learning

“Science, my boy, is made up of mistakes, but they are mistakes which it is useful to make, because they lead little by little to the truth.”
— Jules Verne, Journey to The Center of The Earth

In this chapter we examine several concepts related to local feature descriptor design— namely local patterns, shapes, spectra, distance functions, classification, matching, and object recognition. The main focus is local feature metrics, as shown in Figure 4-1. This discussion follows the general vision taxonomy that will be presented in Chapter 5, and includes key fundamentals for understanding interest point detectors and feature descriptors, as will be surveyed in Chapter 6, including selected concepts common to both detector and descriptor methods. Note that the opportunity always exists to modify as well as mix and match detectors and descriptors to achieve the best results.

Figure 4-1. Various stages in the vision pipeline; this chapter will focus on local feature metrics and classification and learning

Local Features

We focus on the design of local feature descriptors and how they are used in training, classification, and machine learning. The discussion follows the feature taxonomy as is presented in Chapter 5 and as is illustrated in Figure 5-1. The main elements are: (1) shape (for example, rectangle or circle); (2) pattern (either dense sampling or sparse sampling); and (3) spectra (binary values, scalars, sparse codes, or other values). A dense patterned feature will use each pixel in the local region, such as each pixel in a rectangle, while a sparse feature will use only selected pixels from the region.

In addition to the many approaches to shape and pattern, there are numerous approaches taken for the spectra, ranging from gradient-based patch methods to sparse local binary pattern methods. The main topics covered here include:

  • Detectors, used to locate interesting features in the image.
  • Descriptors, used to describe the regions surrounding interesting features.
  • Descriptor attributes, such as feature robustness and invariance.
  • Classification, used to create databases of features and optimal feature matching.
  • Recognition, used to match detected features in target images...