Surrounding Pixel Algorithm

[andrewprivate]

Hello!

I am trying to learn about image comparison algorithms, and I have a question about the surrounding pixel comparison algorithm in this library.

rembrandt/lib/image-comparator.js

Lines 167 to 175 in e402cb4

	let newColorA = this._imageA.getColorAt(currentX, currentY)
	let newDeltaA = this._calculateColorDelta(colorA, newColorA)
	let newColorB = this._imageB.getColorAt(currentX, currentY)
	let newDeltaB = this._calculateColorDelta(colorA, newColorB)
	if ((Math.abs(newDeltaB - newDeltaA) < maxDelta) && (newDeltaA > maxDelta)) {
	return true
	}

Particularly, I am very confused why newDeltaB is subtracted by newDeltaA to compare with the maxDelta. I feel that this may result in a false positive even with very different colors since the differences may cancel each other out in unexpected ways (I cannot really explain this well, but it is rooted in math). Also, the algorithm seems to be asymmetrical, meaning that if I switched the order of the images around, so that imageA became imageB and imageB became imageA, the results would differ greatly.

As a solution to this problem, I thought of another algorithm that could possibly work. Why wouldn't this code be a better alternative?

 let newColorA = this._imageA.getColorAt(currentX, currentY) 
 let newDeltaA = this._calculateColorDelta(colorB, newColorA) 
  
 let newColorB = this._imageB.getColorAt(currentX, currentY) 
 let newDeltaB = this._calculateColorDelta(colorA, newColorB) 
  
 if (newDeltaA < maxDelta || newDeltaB < maxDelta) { 
   return true 
 }