Possible Duplicate:
Recovering data points from an image

Suppose you have a colored plot of several time series (say, from a paper), but not the raw data. How would you estimate the series, which may intersect, from the plot using the axis labels? Bonus cases to consider:

  • Compression artifacts.
  • Semilog or log-log plots.
  • Textured rather than colored series.

Here is a nice example to show the difficulty presented by overlapping, but without any artifacts: MATLAB example http://www.mathworks.com/help/toolbox/econ/multiple_time_series_cointegration_forecast_path.png

If you can solve the base case, feel free to recompress it to introduce artifacts.

  • $\begingroup$ Related: mathematica.stackexchange.com/questions/3831/… $\endgroup$ – celtschk Jun 2 '12 at 19:21
  • 3
    $\begingroup$ Can you upload an example time series figure, some code of what you've tried, etc., and how (specifically) Sjoerd's answer to the above linked question falls short in what you want to do? $\endgroup$ – rm -rf Jun 2 '12 at 19:32
  • $\begingroup$ Here is another solution: mathematica.stackexchange.com/questions/1524/… $\endgroup$ – Ajasja Jun 2 '12 at 19:59
  • $\begingroup$ celtschk's post concerns a different question but @Ajasja's is relevant, however it treats a toy case. I want to consider problems that occur in practice. $\endgroup$ – Emre Jun 2 '12 at 20:15
  • $\begingroup$ Is this in fact a Mathematica Plot? Or is this an image? $\endgroup$ – R Hall Jun 3 '12 at 11:44