The border finder facilitates the finding of frames in paintings or other images. It achieves this by seeding a set of points around the edge of a screen-sized version of the image. It begins to converge these points to another specified point (usually the centre of the image). If the points hit a region of high contrast (in the brightness value), they stop moving. It's like a snake, but it's not a snake.
The border then consists of a set of points, as shown in the example below.
The border is subsampled to a very small image of size 16x16 pixels, which is used as the input to the neural network, which has been trained to recognise shapes of paintings: Square, Circle, Diamond, Rounded Arch or Triptic.
The neural-tester application allows us to view the input alongside the output confidences. So the output, based on the border finder of the above image, gives 77% confidence that it is a triptic painting - which is a fairly high confidence - and with good reason. It's correct!
