Computer Vision News - August 2017

Computer Vision News

To summarize, the main advantages and contributions of GSM are: (1) Improved

real-time matching based on rejection of false matches, by operationalizing

statistical-based motion smoothness properties. (2) Real-time grid-based

approach. (3) approach manages to match features in previously intractable

scenes. (4) Significantly better performance than traditional SIFT, SURF and LIFT

[see the

Given a pair of images taken from different views of the same 3D scene, a feature

correspondence implies a pixel in one image is identified as the same point in the

other image. If the motion is smooth, neighboring pixels and features move

together. This allows to make the following assumption that motion smoothness

will ensure that a tight neighborhood around a true match is the same 3D

location in both images, while the tight neighborhood around a false match are

two different 3D locations. Since true match neighborhoods are the same

location -- there should be many similar features in the surrounding area

(illustrate with yellow circles images below) which means many supporting

matches in the neighborhood. Conversely, false matches should have significantly

fewer supporting matches (red circles in both images). The authors operationalize

this idea into the real-time matching algorithm described next.

Method:

Before diving into the algorithm let’s define the notation we’ll be using: Image

pair

have

{ ,

}

features respectively;

. . .

is the set

of all nearest-neighbor feature matches between and

Images and

are

both divided into G=(20x20) cell grids as illustrated above. We enumerate the

cells of image as and the cells of image as .

is the number of

matches between cells

{ ,

Research

Research