The empty tile threshold is wrong

[raacampbell]

Just looking at fm_167 is obvious this threshold is wrong. I will need it soon for the comb correction so ought to go back and get this working well. All tiles are larger than the threshold. I don't even understand how that happens.

bottomFivePercent = mu(1:round(length(mu)*0.05));
if std(bottomFivePercent)>0.6
  fprintf('%s - Empty tile threshold not trustworthy (STD=%0.2f), setting to mean of dimmest tile.\n',...
   mfilename, std(bottomFivePercent))
  emptyTileThresh=mu(1);
else
  STDvals=zeros(size(mu));
  for ii=1:length(mu)
       STDvals(ii)=std(mu(1:ii));
  end
  f=find(STDvals<0.085); %A threshold that we'll consider to represent the empty tiles
  emptyTileThresh=mu(f(end))*1.01;
end

This is currently done by writeTileStats in the private directory in preProcessTiles. emptyTileThresh is currently only used by calcAverageMatFiles.

In any case, we might well have sections with no empty tiles because of autoROI.

One way that could work is to take a histogram of the tile intensities. Take the mode. Then add a bin or two and call that the the empty tile threshold. This will work for sections with enough tiles. Remember we may have sections with no empty tiles because of the autoROI. Cases where the user changes wavelength during an acquisition might also throw things off.

[raacampbell]

Use a gmm? This is working well for the autoROI.