Bookmark and Share

Computer Vision Metrics: Chapter Seven (Part C)

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 B of Chapter Seven, please click here.

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


Synthetic Feature Alphabets

In this section, we create synthetic ground truth datasets for interest point algorithm analysis. We create alphabets of synthetic interest points and synthetic corner points. The alphabets are synthetic, meaning that each element is designed to perfectly represent chosen binary patterns, including points, lines, contours, and edges.

Various pixel widths or thickness are used for the alphabet characters to measure fine and coarse feature detection. Each pattern is registered at known pixel coordinates on a grid in the images to allow for detection accuracy to be measured. The datasets are designed to enable comparison between human interest point perception and machine vision interest point detectors.

Here is a high-level description of each synthetic alphabet dataset:

  • Synthetic Interest Point Alphabet. Contains points such as boxes, triangles, circle, half boxes, half triangles, half circles, edges, and contours.
  • Synthetic Corner Point Alphabet. Contains several types of corners and multi-corners at different pixel thickness.
  • Natural images overlaid with synthetic alphabets. Contains both black and white versions of the interest points and corners overlaid on natural images.

Note: the complete set of ground truth data is available in Appendix A.

Analysis is provided in Appendix A, which includes running ten detectors against the datasets. The detectors are implemented in OpenCV, including SIFT, SURF, ORB, BRISK, HARRIS, GFFT, FAST9, SIMPLE BLOB, MSER, and STAR. Note that the methods such as SIFT, SURF, and ORB provide both an interest point detector and a feature descriptor implementation. We are only concerned with the interest point detector portion of each method for the analysis, not the feature descriptor.

The idea of using synthetic image alphabets is not new. As shown in Figure 7-2, Rosin[61] devised a synthetic set of gray corner points and corresponding measurement methods for the purpose of quantifying corner properties via attributes such as bluntness or shape of apex, boundary shape of cusps, contrast, orientation, and subtended angle of the corner. However, the synthetic interest point and corner alphabets in this work are developed to address a different set of goals, discussed next.

Goals for the Synthetic Dataset

The goals and...