When I import a PDF file inside Overleaf the math mode characters turns "bold" (shadow)

Question

I used the website diagrams.net to create a figure with some mathematical expressions. Of course, I can export it how PNG and import it to my Overleaf, but I want to retain the vectorization of the expressions. Because of that, I am trying to import it how PDF inside my Overleaf document.

When I use:

\begin{figure}[tbp!]
    \centering
    \includegraphics[width=\linewidth]{images/math_structure.pdf}
    \caption{My figure description.}
    \label{fig:math_structure}
\end{figure}

My figure is shown normally, aparently, but when I zoom in the mathematical expressions I have it:

Image with bold from Overleaf, Firefox and Chrome

Another interesting thing I noted is that when I download the PDF from Overleaf and open it using MUPDF the "bold" disappears, but when I open it using Google Chrome or Firefox the "bold" is there yet.

Image without bold from MUPDF

This is a pretty strange thing because I guess it was a problem of embedding font inside the PDF, but my file opens normally in MUPDF. Does anyone know what is happening and how can I resolve it?

I am sharing the math_structure in order to reproduce the problem in the following link: PDF

Can you specify where you think you see bold text? There is a bit of a shadow around the letters in the first screenshot, but I don't see anything bold. — samcarter_is_at_topanswers.xyz, Dec 21 '21 at 20:30
Can you share the file `math_structure.pdf`. The shadows might be semi-transparent which might not work in mupdf. — samcarter_is_at_topanswers.xyz, Dec 21 '21 at 20:33
@samcarter_is_at_topanswers.xyz Sorry for the delay to answer your questions. Yes, I was referring to these shadows in the letters, because of that I used the expression "bold". I did not know what is it. About the share of the document, I edited the original post with the link to the math_structure.pdf. — Yafrag, Dec 23 '21 at 14:55
Thanks for sharing the file. I just tested in a new project on overleaf, but can't reproduce your shadow https://www.overleaf.com/read/fvjccwvpqmmn https://i.stack.imgur.com/tLiUi.png This might rather be an issue of your web browser than of overleaf. — samcarter_is_at_topanswers.xyz, Dec 23 '21 at 15:15
You could avoid the whole problem by creating your diagram with tikz instead of including an external image — samcarter_is_at_topanswers.xyz, Dec 23 '21 at 15:31
Thanks for your help, I already used tikz in electric circuits diagrams, but I did not think to use it here. Thinking well seems to be the best way to prevent future problems. — Yafrag, Dec 23 '21 at 17:34

score 2 · Answer 1 · answered Dec 25 '21 at 11:44

As an addendum to K J's answer:

looking at each letter there are two objects so although I can not see the shadow within the editor but accept it is there so it must be placed by the text outline generator? here I have moved and coloured some glyphs so the second edge is deliberate but most viewers would not show them as a "GLOW"

Indeed, all those items with a glow are drawn twice in the content stream, once for filling the defining path, once for stroking. E.g. the capital S of "State":

.111484379 0 0 -.111468516 140.764496 314.20746 cm
/G3 gs
55 507 m
55 562.33331 74 609 112 647 c
150 685 193.66666 704 243 704 c
257 704 l
313.66666 704 363 683 405 641 c
426 672 l
429.33334 676.66669 432.66666 681.66669 436 687 c
439.33334 692.33331 442.66666 696.66669 446 700 c
449 704 l
449.66666 704 451 704 453 704 c
455 704 457 704.33331 459 705 c
463 705 l
465 705 468 703 472 699 c
472 462 l
466 456 l
448 456 l
440.66666 456 436.33334 457 435 459 c
433.66666 461 432 467.66666 430 479 c
418.66666 563 385 618.66669 329 646 c
304.33334 656.66669 279.33334 662 254 662 c
218.66666 662 190 650 168 626 c
146 602 135 574 135 542 c
135 519.33331 140.666672 498.66666 152 480 c
163.333328 461.33334 179.33333 446.33334 200 435 c
206.66667 432.33334 235.33333 424.66666 286 412 c
336.66666 399.33334 364.66666 391.66666 370 389 c
408 374.33334 439 349.33334 463 314 c
487 278.66666 499.33334 237.66667 500 191 c
500 137 482.66666 88.333336 448 45 c
413.33334 1.66666412 364.33334 -20.333334 301 -21 c
263.66666 -21 230.33333 -15.333334 201 -4 c
171.66667 7.333334 151.333328 17.666666 140 27 c
122 41 l
119.333336 37.666668 114.333336 31 107 21 c
99.666664 11 93 1.66666698 87 -7 c
81 -15.666667 78 -20.333334 78 -21 c
76.666664 -21.666666 73.333336 -22 68 -22 c
64 -22 l
62 -22 59 -20 55 -16 c
55 101 l
55 180.33334 55.333332 220.66667 56 222 c
57.333332 225.33333 64 227 76 227 c
89 227 l
93 223 95 218.66667 95 214 c
95 192.66667 98.333336 171.66667 105 151 c
111.666664 130.333328 123 110 139 90 c
155 70 177 54 205 42 c
233 30 266.33334 24 305 24 c
336.33334 24 363.33334 36.666664 386 62 c
408.66666 87.333336 420 118.333328 420 155 c
420 183.66667 412.66666 209.66667 398 233 c
383.33334 256.33334 364 272.33334 340 281 c
302.66666 290.33334 278 296.66666 266 300 c
262.66666 300.66666 253.66667 302.66666 239 306 c
224.33333 309.33334 213.33333 312 206 314 c
198.66667 316 188 319.66666 174 325 c
160 330.33334 149 336.33334 141 343 c
133 349.66666 123.333336 357.66666 112 367 c
100.666664 376.33334 91.666664 388 85 402 c
65 434.66666 55 469.66666 55 507 c
h
f
/G7 gs
55 507 m
55 562.33331 74 609 112 647 c
150 685 193.66666 704 243 704 c
257 704 l
313.66666 704 363 683 405 641 c
426 672 l
429.33334 676.66669 432.66666 681.66669 436 687 c
439.33334 692.33331 442.66666 696.66669 446 700 c
449 704 l
449.66666 704 451 704 453 704 c
455 704 457 704.33331 459 705 c
463 705 l
465 705 468 703 472 699 c
472 462 l
466 456 l
448 456 l
440.66666 456 436.33334 457 435 459 c
433.66666 461 432 467.66666 430 479 c
418.66666 563 385 618.66669 329 646 c
304.33334 656.66669 279.33334 662 254 662 c
218.66666 662 190 650 168 626 c
146 602 135 574 135 542 c
135 519.33331 140.666672 498.66666 152 480 c
163.333328 461.33334 179.33333 446.33334 200 435 c
206.66667 432.33334 235.33333 424.66666 286 412 c
336.66666 399.33334 364.66666 391.66666 370 389 c
408 374.33334 439 349.33334 463 314 c
487 278.66666 499.33334 237.66667 500 191 c
500 137 482.66666 88.333336 448 45 c
413.33334 1.66666412 364.33334 -20.333334 301 -21 c
263.66666 -21 230.33333 -15.333334 201 -4 c
171.66667 7.333334 151.333328 17.666666 140 27 c
122 41 l
119.333336 37.666668 114.333336 31 107 21 c
99.666664 11 93 1.66666698 87 -7 c
81 -15.666667 78 -20.333334 78 -21 c
76.666664 -21.666666 73.333336 -22 68 -22 c
64 -22 l
62 -22 59 -20 55 -16 c
55 101 l
55 180.33334 55.333332 220.66667 56 222 c
57.333332 225.33333 64 227 76 227 c
89 227 l
93 223 95 218.66667 95 214 c
95 192.66667 98.333336 171.66667 105 151 c
111.666664 130.333328 123 110 139 90 c
155 70 177 54 205 42 c
233 30 266.33334 24 305 24 c
336.33334 24 363.33334 36.666664 386 62 c
408.66666 87.333336 420 118.333328 420 155 c
420 183.66667 412.66666 209.66667 398 233 c
383.33334 256.33334 364 272.33334 340 281 c
302.66666 290.33334 278 296.66666 266 300 c
262.66666 300.66666 253.66667 302.66666 239 306 c
224.33333 309.33334 213.33333 312 206 314 c
198.66667 316 188 319.66666 174 325 c
160 330.33334 149 336.33334 141 343 c
133 349.66666 123.333336 357.66666 112 367 c
100.666664 376.33334 91.666664 388 85 402 c
65 434.66666 55 469.66666 55 507 c
h
S

The filled version is drawn with the extended graphics state G3, the stroked version is drawn with the extended graphics state G7.

G3 fills in an opaque manner:

<</BM/Normal/ca 1>

but G7 strokes very transparently (opacity .1098) and sets some other parameters:

<</BM/Normal/CA .1098/LC 0/LJ 0/LW 0/ML 4/SA true/ca .1098>>

But in particular G7 also sets the line width to 0 (the thinnest line that can be rendered at device resolution: 1 device pixel wide).

The OP mentions that they see the shadows when they zoom in. Thus, maybe those viewers in which you see a broad shadow/glow after zooming do simply zoom by drawing everything magnified by the zoom factor, i.e. the shadow/glow becomes zoom factor * 1 pixel wide; and those viewers in which you don't see a broad shadow/glow draw the outlines even after zooming with a 1 pixel width.

*"some increase in anti-alias depth"* - that also might be true as also **SA** is set to **true** in **G7**. Also possible is missing support of some entries in the graphics state. Without access to the viewer in question this essentially is guesswork. With access one could run some test pdfs through it... — mkl, Dec 26 '21 at 09:44

K J · Answer 2 · 2021-12-25T22:27:20.103

It does not appear to be the difference is in the font style since the weighting between standard 24 and bold 24 is shown below on the right. Which is not evident in your two samples.

However, what is noticeable in your sample is the "shadows" around each of those letter on the left giving the impression of extra thickness.

Initially I would expect that could be caused by the difference between jpeg (haloed lettering) and png (crisp anti-alias outlines). But then the shadow is too regular i.e. not uneven like it would normally be in a jpeg.

At this stage it looks like there may be some other reason for such fuzzy fonts. Without a sample I would have to guess the PDF has potentially a font with an alpha component but could be way off in such a wild assumption.

Later Edit

Thanks for your link but the mystery deepens, since that linked PDF in Chromium Edge even enlarged shows no evidence of any shadows, but then again the maths looks like vector outlines only the middle Tahoma appears to be font and the one embedded, as generated by Skia/PDF thus built by chrome?.

I have to agree there is some other influence somewhere down the line but the browser should not affect the PDF unless it adds or respects some overlay based on an extra component, and looking at each letter there are two objects so although I can not see the shadow within the editor but accept it is there so it must be placed by the text outline generator? here I have moved and coloured some glyphs so the second edge is deliberate but most viewers would not show them as a "GLOW"

You mentioned "diagrams.net" which does have many shadow options but I never experienced any other than deliberately set to right and down. Perhaps look for a rogue setting there.

In summary the file is declared as compatible with version 1.4 (may have transparency) and clearly some transparent objects have been included around each letter! ~~but not in a fashion expected by all viewers.~~ As a result of @mkl 's observation I retested the pdf in many viewers with the settings that could have an effect such as vector line thickening in acrobat, However NONE I tested showed the extra thick outlines, thus the PDF seems valid but some PDF viewer(apps) methods you are using seem to thicken the anti-alias much more than should be expected for a single pixel boundary.

Makes sense what you said, I guess your assumption is correct. However, I need to understand how can I set this alpha to 0 by default. Because I can not bet that the reader opens my document with a specific set of PDF reader software. Or in the worst case, the printer prints the shadows. — Yafrag, Dec 23 '21 at 15:25

When I import a PDF file inside Overleaf the math mode characters turns "bold" (shadow)

2 Answers2