Split Filter Contours

[osulacam]

Adds best fit bounding boxes and splits filter contours into two operations.

[codecov-io]

Current coverage is 57.85% (diff: 44.57%)

Merging #627 into master will decrease coverage by 0.10%

@@             master       #627   diff @@
==========================================
  Files           196        197     +1   
  Lines          6165       6238    +73   
  Methods           0          0          
  Messages          0          0          
  Branches        561        566     +5   
==========================================
+ Hits           3573       3609    +36   
- Misses         2427       2464    +37   
  Partials        165        165          

Powered by Codecov. Last update a700892...394cd58

[SamCarlberg]

Nice. Though might it be a good a idea to use best-fit bounding boxes instead of the simple x-y-oriented bounding boxes?

Take a look at this image. The current bounding box we use to calculate the ratio will be the blue rectangle, which has a very different ratio than the object (the oblong rectangle)

[osulacam]

Best fit bounding boxes are useful but may not be what all users want. There could be an option for it but that might complicate the step.

[SamCarlberg]

Yeah, it would be hard to change the existing stuff because it's already in use. I'm reluctant to add more stuff to the filter operation because there's just so much crap packed into it already and it makes the UI a mess. @JLLeitschuh what are your thoughts?

[JLLeitschuh]

Uhhhhh... An AdvancedFilterArgs object that you can pass in?
Just a thought.

[osulacam]

We could just make a separate advanced filter contours and simple filter contours operation. The simple would just have area and perimeter, while the advanced could have everything.

[osulacam]

Bytedeco RotatedRect is broken so we can't actually do the best fit bounding boxes.

[SamCarlberg]

How are they broken? I got them working fine when I was playing around with them.

[osulacam]

They don't give a list of points for the vertices and the bounding rectangle wasn't even remotely correct so I couldn't use that either.

[SamCarlberg]

RotatedRect contains the center point, width, height, and rotation. You can calculate the vertices from that. And how was it so far off? They've always been accurate for me. I assume you're using something like

for (int i = 0; i < contours.size(); i++) {
  RotatedRect rr = opencv_imgproc.minAreaRect(contours.get(i));
  // ... 
}

Edit: take a look here

[SamCarlberg]

Get rid of the space in Math.sqrt( Math.pow.... Should also use a sensical scheme of line breaks, right now it's hard to keep track of what's going on. I'd put the Math.pow statements each on their own line.

[SamCarlberg]

This seems like a breaking change. Old FilterContours steps won't be compatible.

[osulacam]

The integer division problem was fixed by #618. The rest of this PR will split up filter contours to make it easier to use.

[AustinShalit]

@osulacam Is this PR ready for review?

[osulacam]

@AustinShalit Yes

[SamCarlberg]

I'll take a look at this tonight

[JLLeitschuh]

Is it still a breaking change or has that been resolved?

[osulacam]

Because this technically creates a two new operations to get rid of an old one it has to be a breaking change. It is still a necessary change that shouldn't break anything significantly.