op warpRing
Thos FireSight stage creates recognition templates for different angles using ring projection or a set of specific angles. Template matching is generally sensitive to the angle of the template with respect to the image. However, it is possible to create a rotation-invariant template matching by "stacking and averaging a bunch of rotated templates." A fully rotation-invariant template is created by means of ring projection.
Although rotation invariance is achievable, it comes at a cost. For example, the ring projection of a rectangle template will recognize a "+'or an "X". In other words, there is a tradeoff between rotation-invariance and matching accuracy. One way to address is this is only specify the angles you need. If vertical and horizontal template orientations are the only ones that matter, a [0,90,180,270] multi-angle template will reject an "X".
-
angles JSON array of angles (e.g.,
[0,30,60]. Default is[], which produces a ring projection of input image.
The stage model for warpRing gives the result image size of the generated template, which will vary according to the angles specified. A ring projection will always result in a square image:
{
"width":240,
"height":240
}
Example: 4-way rectangle template pipeline
firesight -i img/rectangle100x162.jpg -p json/warpRing-4way.json -o target/rectangle100x162-4way.jpg
Example: 8-way rectangle template pipeline
firesight -i img/rectangle100x162.jpg -p json/warpRing-8way.json -o target/rectangle100x162-8way.jpg
Example: Ring-projected rectangle template pipeline
firesight -i img/rectangle100x162.jpg -p json/warpRing.json -o target/rectangle100x162-ring.jpg
Example: Spin the duck pipeline
firesight -i img/duck.jpg -p json/warpRing-4way.json -o target/duck-4way.jpg
