Depix is a program to convert plots back into datasets.

Despite the simplicity of Depix, it provides a very convenient work flow for transforming (scanned) plots back into (x,y) datasets.

Depix was created on a standard Linux Desktop environment with Python, Numpy, Gimp, pdfimages, and Inkscape installed. I have only used it on such an environment. If you get it working on another OS, feel free to make a pull request.

If you encounter troubles when using Depix, check the FAQ at the bottom of this file. If that does not solve your problem, feel free to get in touch: [email protected].

• This program was written by Toon Verstraelen in Sep 2012 for the preparation of a paper with the title, "ACKS2: Atom-Condensed Kohn-Sham approximated to second order", in order to convert some plots of an old JCP paper into datasets for comparison with a newly proposed model.

• Support for logarithmic scales was added May 2016.

1. If the plot is part of a pdf, use pdfimages to extract the figures from the PDF as .pbm files. One can also use Evince to extract bitmaps from a PDF: right click on the image and "Save Image As ...".

2. Open the image containing the plot with Gimp, crop it if needed and save the file as a .png file.

3. Open the .png file with Inkscape. Use the "link" option to avoid large .svg files in one of the following steps.

4. Draw the x- and y-axis as two separate straight line segments, i.e. lines with only a beginning and end node. The accuracy of the extracted data will improve when these line segments are made as long as possible. However, make sure the beginning and the end node lie at sensible points on the x- and y-axis. Depix can handle cases where the x- and y-axis are not orthogonal or rotated, e.g. due to a poor scan.

5. Draw a polyline, consisting of straight line segments (not Bezier curves), over the curve of interest. Make sure your drawing falls nicely over the curve in the scanned image, as this will also determine the accuracy of the final data. It may be helpful to use a bright colored and semi-transparent line style.

6. Open the XML window in Inkscape. Select the x-axis and change the id of the path to xaxis:x0:x1:kind, where x0 and x1 are replaced by the numerical x-values of the beginning and end points of the line segment for the x-axis. The last part, kind, must be replaced by lin or log for linear or logarithmic scales, respectively. Do the same for the y-axis, using the id yaxis:y0:y1:kind, following the same conventions. The path with the datapoints should be given the id "data".

7. Use the "File" -> "Save as" menu item to save the file in .svg format.

8. Run Depix as follows:

    ./depix.py yourfile.svg yourdata.txt
   

   Some screen output is given, which may be useful for debugging, if things don't work as expected. The text file yourdata.txt contains two columns with x and y data.

This work flow has the advantage that all intermediate steps are saved in files, such that you can make incremental improvements. Furthermore, thanks to the neat graphical interface of Inkscape, one can make fairly accurate overlays of the data points on top of the scanned figure.

If you need to extract more than one curve from a single plot, just copy the .svg file with the first curve to a new file and trace the second curve in this copy.

The example hf_135.svgcomes from the paper http://dx.doi.org/10.1063/1.466016. The figure is converted to a dataset as follows:

./depix.py hf_135.svg hf_135.txt

The example diffusion.svg comes from the paper http://dx.doi.org/10.1016/j.molliq.2014.11.010. figure is converted to a dataset as follows:

./depix.py diffusion.svg diffusion.txt

Q. I get errors when running Depix, e.g. like the following:

Traceback (most recent call last):
  File "./depix.py", line 177, in <module>
    main()
  File "./depix.py", line 170, in main
    data = process_file(args[0])
  File "./depix.py", line 159, in process_file
    x_axis, y_axis, px_data = load_pix_data_svg(fn)
  File "./depix.py", line 76, in load_pix_data_svg
    px_data = parse_data(path)
  File "./depix.py", line 56, in parse_data
    x, y = word.split(',')
ValueError: need more than 1 value to unpack

A. Make sure the x- and y-axis are straight lines and the data polyline consists of straight line segments. This may happen when Bezier curves were used instead.

Q. The data points are mirrored horizontally or vertically. How can I fix this?

A. One must make sure that the order of the two points in the line segments for the x- and y-axis are compatible with the order of x0 and x1 in the path id xaxis:x0:x1:kind. One may add an end marker in Inkscape to see clearly what the end point of the line segment is. With the menu item "Path" -> "Reverse path", one can swap begin and end.