While these techniques construct feature detectors that optimize for repeatability across all ob-ject classes, it is also possible to develop class- or instance-specific feature detectors that maximize …

The key to feature detection and matching is to find repeatable features and distinctive descriptors that are invariant to geometric and photometric transformations.

Feature Descriptors Once we have detected distinctive and repeatable features, still have to match them across images – Image alignment (e.g., mosaics), 3D reconstruction, motion tracking, object …

Many of these descriptors also come with a feature detector, which is typically a more sophisticated version of the Harris corners we saw earlier in this document.

Section II provides an overview of the recent state-of-the-art feature detection and description algorithms proposed in literature. It also summaries and compares their performance and accuracy under …

To get more accurate real-number coordinates, need to run subpixel algorithm. General idea: starting with an approximate location of a corner, find the accurate location that lies at the intersections of …

Group all features in the training set into a finite number of clusters (K-means clustering, GMM clustering, etc.), and represent each cluster by a “mean feature vector” (a visual word)