Why am I requested to compile my figure files with 300 DPI?

Question

I will submit an article to Geoscientific Model Development, and the submission instructions requests that I ensure that my figure files are "compiled as *.pdf, *.ps, *.eps, *.jpg, *.png, or *.tif files with a resolution of 300 dpi" before I submit the manuscript for peer review.

What does it mean in practice that my figure files are compiled with a resolution of 300 DPI? How is this relevant when it comes to PNG and PDF files? Do the DPIs of PNG and PDF files have any effect on the final document, and is this even information that is stored in such files?

I know what DPI is and the difference between a raster and a vector image. That's not the issue.

Answer 1

Well, the instructions are a bit old-school. They have the underlying assumption that you submit your figure files in exactly the size they are printed and they want the resolution of any pixel content of the final paper to be 300 dpi. Of course, in modern typesetting, you can embed any figure at any size. So the instructions translate to: If your figures are embedded at a reasonable size in the final paper, any pixel content should be 300 dpi.

All of this in turn hardly ever matters if you follow the other guideline, which says:

The first choice should be vector graphics

… and for vector graphics, you do not have pixel content in the first place. So, if this is about plots, diagrams, etc., just deliver them in a vector format.

Now, depending on the subject of your research, you may have photographs and similar, but here you cannot control the resolution at this point and are stuck with what you have (see: Is increasing the DPI of a low resolution image a good idea?).

The only case in which you should worry about this guideline is if you have some content that needs to be rendered to pixels and you can control the resolution of the rendering.

Answer 2

PNG files are pixel graphics. The requirement simply says that you need to include pictures in PNG format that have a sufficiently high resolution so that they don't look "pixely". In other words, you shouldn't include a 100x100 pixel picture and display it in a figure that covers half a page. (But if that is an inset that shows up in a 5x5 mm area, then 100x100 pixels are enough because that equates to ~500 DPI.)

Answer 3

DPI matters for raster graphics, which encode individual pixels. This is to be contrasted with vector graphics file which contain information about shapes. Some formats, like PDF, can contain both raster and vector graphics. PNG is purely a raster graphics format. Too low DPI for raster graphics will look bad if the journal increases the size of the figure. 300 DPI is a common minimum guideline for raster images, but it's generally better to create vector graphics whenever possible.

Answer 4

In some way or other, you will not only submit your images (say, a PNG graphic that is 600 pixels wide and 200 pixels high), but also layout "instructions" where and how big to put that image on a page of (ultimately) paper. If these instructions amount to "as wide as two text columns", then those text columns better be no wider than one inch each, because at 300 dpi, your 600 pixels are good only for 2 inches. If the numbers don't match, your options are to decide that the image shall appear smaller in size (and thereby less impressive, or perhaps your important 3-by-3 pixel markings become almost invisible to the naked eye) or you try to render it at higher pixel resolution.

Answer 5

1. Prefer vector graphics. If you plot using e.g. Matplotlib, then simply outputting a PDF file does the trick in most cases:

   fig, ax = plt.subplots()
   ...
   fig.savefig("figure.pdf")
   

2. If it must be a PNG file, ensure that it has sufficient pixels to meet the target print DPI and print width. That is, given:

   # An 8 inch wide figure at 300 DPI requires 2400 pixels.
   print_dpi = 300
   print_width_inches = 8
   print_width_pixels = print_width_inches * print_dpi
   Your PNG file.
   file_dpi = ...
   file_width_inches = ...
   file_width_pixels = file_width_inches * file_dpi
   

   The following condition must be satisfied:

   assert file_width_pixels >= print_width_pixels
   

   What really matters is if you have the number of pixels needed, file_width_pixels (e.g. 2400 pixels wide). As long as you create a figure with enough pixels, you are good. The other settings (file_dpi, file_width_inches) can be chosen arbitrarily to match this constraint, and don't really matter.

   In fact, you can easily convert an image with "1 DPI" in the metadata field into a "9000 DPI" image with the following command:

   exiftool -XResolution=9000 -YResolution=9000 figure.png
   

   This command neither adds nor removes the only things that are actually important: the pixels. For comparison:

   Image A	Image B
   file_dpi = 1	file_dpi = 9000
   file_width_pixels = 234	file_width_pixels = 234

   Clearly, both images are exactly 234 pixels wide, and thus have the exact same quality. The DPI metadata is irrelevant. Look at the number of pixels, not the supposed DPI.

   Regarding Matplotlib again, you may generate the appropriately sized PNG via:

   fig, ax = plt.subplots(figsize=(file_width_inches, file_height_inches), dpi=file_dpi)
   ...
   fig.savefig("figure.png")
   

   Again, the file_dpi does not need to be 300 in order to produce a high quality image. (e.g., file_width_inches=10000000000000, dpi=1 is a perfectly valid choice from the publisher's point of view, though your fonts will be very, very tiny, and your computer may not have enough memory for an image with higher resolution than images from the Hubble Space Telescope. :))

Answer 6

Technically you can rescale.

I always understood this as a hint to check if your figures are going to look as you desire if they are scaled to the planned size and rendered at 300dpi.

That could affect:

• rastered graphics (which you may like to represent in an integer multiple to avoid fuzzy edges)
• minimum line widths/usage of different line widths to indicate data traces in plots
• filling patterns
• marker types which you use in plots
• potentially fonts/sizes which you use (-> font hinting)
• checking if gray values can be represented appropriately by the raster (300dpi is the B/W resolution, not the resolution in which you can represent a gray value)

The other point which I always understood implied was an polite version of "please don't send a 1200x1200 resolution rasterized graphics".

Answer 7

This sounds like more of a question for Google than for Academia Stack Exchange, but since preparing figures for the final manuscript and typesetting is always a hassle (would love to hear from anyone who thinks it isn't), here are my two cents:

1. Anything that is printed will need a higher resolution than something on screen. For example, powerpoint files generally work with a screen resolution of 72 dpi because all they are meant for is to be projected. If you don't want your images to look pixelated in print, you need to ensure they have a resolution of at least 300 dpi. That's just how printing works.

Familiarize yourself with this and differences between vector and raster images - there is lots of information out there online, also through printshops etc. because the same thing holds for printing posters for conferences.

2. In general, what helps is to prepare your figures at the real (i.e. actual, final) print size. That way you can ensure (by doing test prints yourself) that everything is up to par while preparing the figures.

3. Save .png for the web, .jpg for test prints/first submissions and always make sure you have high resolution files (.eps, .tif) for final submission. There will always be loss of data when you export as .png and .jpg because there is compression and loss of data (which is why the file sizes are smaller - and we like that until the publisher informs us that the resolution of our images is too low).

Of course any of this may become obsolete in a paperless world, but this is not the reality we live in so it helps to think of eventual printing at all times and to make sure that your raw files and source material do(es) have the right resolution and/or can be exported as such.