Thanks @DaveTryon I'm going to take a look at this this week.

@DaveTryon I don't think this is the way we should solve this problem, though your test case is super helpful! Would you mind if I pulled your branch down and tweaked some other paramaters and asked you to review them?

The problem your test case exhibits is solved not by "overlay detection" but by our gradient background logic which "should" but obviously isn't handling this scenario. I would prefer to tweak existing parameters to creating a new one. The problem with the logic as you've tweaked it, is that it will filter out many cases that we are currently detecting correctly.

If this is good with you, I'll take your logic and ensure your test case works, but by tweaking some other parameters. If there are any other test cases you want to add before I do this let me know!

@chriscm2006, I agree that this is not an elegant fix. If there's a way to adjust the gradient algorithm to handle this case, then I'm all for that. Please feel free to fork this branch--I'll add a few more images to serve as test cases, just to broaden the data set that gets checked. I'll leave the PR open until I get the chance to add the additional test cases (probably on 1/31)

@chriscm2006, I've added a few more test cases to https://github.com/DaveTryon/axe-android/tree/TextOverlayBugDemo. Please feel free to use these as appropriate to tune the color contrast handling. I noticed that some of the new test cases report low confidence instead of high, so the final form of the test cases may need to be tweaked accordingly. My intent here was to capture some interesting test cases. I'll close out this PR now.