Troubleshooting Some Options for Converting PDF to JPG or Other Image Formats

I had a bunch of PDFs.  I wanted to glance at just the first page of each.  This post describes some steps I took to find a way of extracting those first pages and making them easily available.

I believed it was going to be a simple matter to export from PDF to JPG using either Adobe Acrobat or IrfanView.  I had used both successfully in the past for this kind of purpose.  Unfortunately, for some reason Acrobat's Advanced > Document Processing > Batch Processing option was not working for me on this particular day.  Instead, I went into IrfanView > File > Batch Conversion/Rename, set the output format to JPG, selected the folder containing the newly created PDFs, clicked "Add All," set the output directory, and clicked Start Batch. Yet here, again, unknown causes were conspiring against me. IrfanView said, "Error! Can't load [filename]" for each of the PDFs. I tried exporting to PNG instead of JPG; same result. Was I mistaken in believing that IrfanView could produce JPGs from PDFs?  It had worked in a previous post.  Why not now?

A search led to a thread that raised the question of whether the source files were corrupt. I checked several of these newly created PDFs and they did seem to open OK in Acrobat.  So corruption didn't seem to be the issue.

At first, in that previous post, I had gotten that "Can't load" error.  The solution there was to download an upgrade to Ghostscript. Maybe I had updated IrfanView since then and, in the process, had somehow broken the Ghostscript update? I'd also gotten the error, another time, when trying to use IrfanView for audio conversions, which it apparently couldn't do.  So, ah, maybe what I was trying to do was confusing IrfanView.  I tried again with just one PDF.  There didn't seem to be much that could go wrong with that.  But it still didn't work.  When I tried to open a PDF with IrfanView, it gave me a message:

"Decode error! Can't load Ghostscript or Ghostscript error. Install Ghostscript from http://sourceforge.net/projects/ghostscript or http://sourceforge.net.

I went to the former. It pointed me to another page, and that one pointed me to yet another. It looked like Ghostscript had indeed been updated within the past month or so. I downloaded and installed the update. It didn't solve the problem, though possibly a reboot would have helped.

Was there another way to export JPGs or PNGs from PDFs? I tried XnView, which Irfan claimed was somehow built on code appropriated from him. Its interface certainly was very similar in this particular task. It didn't produce any JPGs either. A post reminded me that I hadn't tried IrfanView in command-line mode. The formula I had worked out previously was like this:

i_view32 D:\Folder1\File35.pdf /c=d:\TestFolder\File35.jpg

but now that was giving me Ghostscript errors too. A search led to suggestions to try ImageMagick and iTextSharp. The latter seemed beyond me. There were also a couple of suggestions on using scripting in GhostScript. They were a bit technical for my taste at this time.

It looked like ImageMagick had been around for a long time -- there were books about it -- and I had run across a number of references to it. So I decided to start there. The copy I downloaded from CNET was corrupted, so I got another from the ImageMagick FTP site. Unzipped, the ImageMagick program folder was 146MB and contained 5,694 files. This was no little alternative to IrfanView. This was a doctoral dissertation.  I immediately cast about for a GUI front end -- I couldn't even find its executables -- and alighted on Converseen.  I downloaded it from Softpedia and installed it.  Would it convert my PDFs to JPGs? I named three PDFs to try it. All three failed with this error:

Error: Magick: Postscript delegate failed [filename]: No such file or directory @ error/pdf.c/ReadPDFImage/664

Now, what do you suppose that meant? I tried a search and got a couple of very helpful sites in Chinese. I modified the search and, lo, it looked like that was an ImageMagick error, not a Converseen error. I gathered there could be a couple of possibilities. One was that, as someone reported, this error was likely (for some unknown reason) when converting PDFs that were mostly text, as mine were, rather than mostly image. Also, there seemed to be a regular cascade of programs: Converseen was a front end for ImageMagick which used GhostScript. So I couldn't necessarily be sure which one was responsible. For that matter, I didn't even know where to put ImageMagick; I wasn't seeing a manual. Further reading suggested that resolving problems in these programs could be challenging.

I tried another search. At the top of the list: Boxoft PDF to JPG Converter. Its installation process said, "Requires Ghostscript. Do you want to get a free GPL Ghostscript?" It seemed I was destined to have a computer full of copies of Ghostscript.  Or maybe this installation would be the answer to all of the problems described above.  I downloaded and installed the proffered Ghostscript 8.71.  The process didn't look familiar.  I wasn't sure what I had been doing with Ghostscript previously, but this wasn't it.  So would this perchance fix IrfanView?  I opened IrfanView and, wow, now it ran, no reboot necessary.  The way to fix IrfanView was to install Boxoft (or, possibly, to reinstall IrfanView).

But something was wrong.  Didn't IrfanView normally give me multiple pages, one PDF per page, when I used it to convert PDFs to JPGs?  This time, I was getting only one JPG per PDF, no matter how many pages it had.  I assumed that later Ghostscript installations replaced earlier ones, but now I saw that was not necessarily so.  I went into Control Panel > Programs and Features.  I saw three different Ghostscript items.  I wasn't sure if I could uninstall two of them without screwing up the third one.  I tried it, keeping only the Ghostscript 8.71 that I had just installed.  Boxoft and IrfanView both ran as before.  So cleanup accomplished, but IrfanView problem not fixed.

Since I now had Boxoft installed, I tried that.  Like IrfanView, it had a command-line mode and a GUI mode.  I tried the GUI.  It had multiple options.  I liked it.  It did produce multiple JPGs per PDF.  They looked good.  I ran it again, this time selecting only page 1 (Settings > Common tab > PDF Convert Range), and that's exactly what it gave me.  So I wouldn't have to take a separate step of deleting JPGs for the pages other than page 1 of each PDF.  IrfanView didn't have that option, though it did have image editing options (if e.g., I wanted to change the dots per inch, size, or coloring of the resulting JPGs).

I could have continued to look at other options appearing in that search, but it appeared a working conclusion was to go with Boxoft and/or IrfanView, depending on the kind of output desired, and to make sure I was using just one stable, recent version of Ghostscript.

Using Screenshots to Examine Many Files Quickly

I had a couple of projects that called for a way to examine a large number of files.  It seemed that screenshots could help in those projects.  This post describes the techniques I used.

EML Analysis

In one project, I was working with various email files that I had exported from Thunderbird.  These files had an EML extension.  Typically, if I viewed an EML file in Notepad, I would see various codes and other information that wouldn't be visible if I viewed it in an email program like Thunderbird.

I was interested in seeing the header codes in these EML files.  Those codes appeared at the tops of the files.  I felt that I could probably see what I needed to see in the first screenful of a Notepad session, opened maximized.

In other words, the concept was that I would open the EML file in Notepad; I would take a screenshot; I would save the screenshot; and then I would close the file and repeat the process with the next EML file on my list.  Then I would combine all those screenshots into one file, and flip through it or perhaps use other tools to analyze it further.  I wouldn't have to sit there, maintaining constant attention while the process continued in real time; I could just review the outcome afterwards.  (For some purposes, an alternative would have been to combine or select from the text, without a graphical view.)

The first step was to build the list of EML files that I wanted to examine.  I moved them all into a single folder and used DIR and Excel to give me the list and to convert it into a series of batch commands.  There was one such command for each such EML file.  Before running those commands, I had to open Notepad once, turn on its Format > Word Wrap option, and then close it.  The format of the command was as follows:

start /max notepad "D:\Folder Name\Email Name.eml"

That command was sufficient to open the EML file.  Next, I needed to pause the system for a moment, so that the file would have time to come onscreen. Among numerous suggestions, I favored a command involving PING ("ping 1.1.1.1 -n 1 -w 1500 > nul") because of its fine-tunable setting (in the example just given, 1500 milliseconds).  Unfortunately, that command's output component (" > nul") would have prevented me from adding more commands on the same line.  So I had to go with "TIMEOUT /T 1" for a one-second delay.

Next, I needed a command to take a snapshot.  It looked like there were multiple options here.  I had already installed NirCmd and had found it useful for other things, so I used this command:

start NirCmd savescreenshot "D:\Folder Name\Screenshots\Email Name.png"

NirCmd came with an option to copy its executable (nircmd.exe) to C:\Windows, so that this command could run without any need to specify the location of NirCmd, to put a copy of it in the current working folder, or to modify the computer's Path.  NirCmd wasn't saving to subfolders properly, so in the end I had to modify that part of the command.

Finally, I needed a command to close Notepad.  The advice that worked for me was:

taskkill /f /im notepad.exe

Note that this would close all currently open Notepad sessions.  These three steps (i.e., open the EML in Notepad, take a picture with NirCmd, close Notepad) would give me a screenshot of the first screenful's worth of the file's contents.  Collectively, those screenshots would give me a visual impression of the various kinds of codes appearing at the start of my EML files.

I used && to combine multiple commands on the same line, as a single (long) batch command. If that had failed, I could have added index columns next to the spreadsheet columns in which I built those two commands, with alternating even and odd numbers in those columns:  1 for the first Notepad command, 2 for the first NirCmd command, 3 for the second Notpad command, and so forth.  These index numbers would allow the various commands to be sorted into proper sequence in a single column, for copying and pasting into a batch file.

In short, for each EML file, I combined four commands with &&, into a single long command like this:

start /max notepad "D:\Folder Name\Email Name.eml" && timeout /t 1 && start NirCmd savescreenshot "Email Name.png" && taskkill /f /im notepad.exe

This gave me some PNGs.  Now there was the question of what to do with them.  One option was to simply stitch them together in a slideshow (using e.g., IrfanView) or a single PDF (using e.g., Acrobat).  I did a brief investigation of OCR software for that purpose.  Ultimately, I just used IrfanView, without even creating a slideshow, to arrow down through those PNGs, one at a time, at whatever pace I chose.  So I could look at whether each page came through OK.

PDF Analysis

In another project, I had a bunch of PDFs that I had created in a conversion process.  I wanted to check if the PDFs came through OK.  It would have been very slow to open them, one at a time, and page through them.  Combining them all into a single large PDF, which I could also page through, would have produced a huge file.  Also, if I was working with large PDFs or many PDFs (or both), I might have to look at huge numbers of pages.  Boredom or haste could lead me to flip past an important one-page document, while checking hundreds or thousands of less important pages.

Based on various factors (including the number of PDFs, their importance, and the time available), I decided to examine just the first page of each PDF.  I might not be able to tell if the whole document printed properly, but at least I could eliminate those instances where printing failed completely.

For this purpose, the process described in the previous section offered one possibility.  I could probably work up a set of commands to open a PDF, take a screenshot, and then close it, and then flip through the resulting screenshots.

I did not actually pursue that approach in this case, however.  Instead, I wanted to see if I could convert the PDF documents to JPG and then flip through just the first page from each such document.  If I had a hundred documents to check, I would have a hundred pages to look at -- not a thousand.  A separate post discusses that investigation.  The tool I chose was  Boxoft PDF to JPG Converter.  Another way to proceed might have been to split the PDFs first, using something like PDFsam, and then combine the PDFs of each resulting first page into a larger PDF that I could flip through.

Scanning Functionality for a Brother MFC-7340 Multifunction Device in Ubuntu 10.04

I had installed the printing functionality for a Brother MFC-7340 multifunction device in Ubuntu 10.04 (Lucid Lynx), including a Windows XP virtual machine running in VMware Workstation 7 on that Ubuntu installation.  Now I wanted to set up scanning functionality as well.  This post describes that effort.

I first verified that the scanning functionality was not installed automatically as part of the printer driver.  I went into Ubuntu’s Applications > Graphics > XSane Image Scanner.  (I was not sure whether XSane came with Ubuntu or whether I installed it separately through Synaptic Package Manager.)  XSane said it was “scanning for devices” and then reported that there were “no devices available.”  It occurred to me that this could be because I had connected the printer to the virtual machine, which probably would have made it unavailable for Ubuntu itself.  I went into the virtual machine and selected VM > Removable Devices > Brother Printer > Disconnect.  Then I went back out to Ubuntu and tried again in XSane.  I got the same result.  So it did seem that I would need to install scanner drivers separately.

The Brother Drivers for Linux webpage did have drivers and instructions for scanners.  I went to the driver download page, where I found that the MFC-7340 was categorized as a “brscan3” model.  I selected the Debian driver for the 32-bit brscan3.  There was also an option to download and install a scan-key-tool.  I wasn’t sure what that was, or if I would need it, so I held off.  The 32-bit brscan3 download gave me a file called brscan3-0.2.9-1.i386.deb.  There were different installation instructions for USB and ethernet connections.  There was also a separate webpage for Scanner Settings for Normal Users.  They had an Ubuntu 10.04 option there.  The steps I had to take there were, first, to type “sudo gedit /lib/udev/rules.d/40-libsane.rules” and then add two lines at the end of the, right before the line that began with “# The following rule will disable USB autosuspend for the device.”  The two lines I had to add were:

# Brother scanners
ATTRS{idVendor}=="04f9", ENV{libsane_matched}="yes"

Then I had to save that file and reboot the system.  When that was done, as with the printer configuration, I went through the Pre-required Procedures that seemed to apply to all versions of Ubuntu generally or to this version (32-bit 10.04) particularly.  Those steps seemed to be as follows:

sudo -i
mkdir /var/spool/lpd
apt-get install sane-utils
apt-get install psutils

I got an error for the mkdir command, as this was a step I had already done – I think when I installed the printer driver.  I already had sane-utils too, and got a message telling me to run “apt-get autoremove” to get rid of some packages that were no longer required, which I did.  I already had psutils as well.  With the Pre-Required Procedures out of the way, I was ready to follow the USB installation instructions for the scanner driver.  Having already downloaded the driver and connected the MFC-7340’s USB cable, I navigated to the folder where I had put the download and then typed this:

dpkg -i --force-all brscan3-0.2.9-1.i386.deb
dpkg -l | grep Brother

This all seemed good.  I started XSane, adjusted its settings, and scanned.  It worked with the automatic sheet feeder.  It saved in multiple formats, including JPG and PDF.  I tried scanning while standing at the scanner and punching the buttons on the MFC-7340.  That didn’t work.  That, then, was what the scan-key-tool driver was for.  From the download page, I got the 32-bit Debian brscan3 scan-key-tool driver (brscan-skey-0.2.1-3.i386.deb) and looked at the installation instructions.  They said I needed GIMP, which I already had installed (I think it came with Ubuntu), and that I also needed to have installed the scanner driver already, which I had done.  That took care of the Pre-Required Procedures.  Now, as above, I navigated to the folder where I had put the download and entered more or less the same commands as with the scanner driver, plus a couple of additional steps:

dpkg -i --force-all brscan-skey-0.2.1-3.i386.deb
dpkg -l | grep Brother
brscan-skey
brscan-skey -l

Following their additional instructions, I set the scan-key-tool to run automatically when I started the computer, by going into System > Preferences > Startup Applications > Add (suggested name = Brother Scan Key, command = brscan-skey, comment = Scan from the MFC-7340 console).  I scanned from the console.  The resulting file was saved in /root/brscan, probably because I had run all of these commands as root (i.e., sudo).  I ran brscan-skey again from my normal user (i.e., not root) prompt.  This time, I saved to Image rather than to File.  This provoked GIMP to start up and display the scan.  I saw that, again, it was saving the scans to /root/brscan (though I could only see them when I went there as root).  I changed the file name and location to a PDF in my preferred folder for scans, but GIMP was not prepared to do PDFs, so that was apparently an advantage of saving as File rather than as Image.  I tried again with GIMP, this time as a JPG to my preferred folder.  That worked, but the scankey was still saving copies of them to /root/brscan too.  The instructions said I could type “brscan-skey -u” to change the target user, so I did that as root, and then typed brscan-skey as myself.  Unfortunately, that didn’t do it; it was still saving in /root.  I tried “sudo brscan-skey -t” to stop it altogether.  That worked.  Now it wouldn’t scan from the console.  I started it as myself again.  Now it was saving to /home/ray/brscan.

The next step I wanted to take, following their additional instructions, was to modify the default script for scan-to-image so that it would save in my preferred folder as a JPG, and scan-to-file so that it would save there as a PDF.  Brother suggested changing this line in the default scantoimage script:

scanimage --device-name "$device" --resolution $resolution > $output_file

to this:

scanimage --device-name "$device" --resolution $resolution | pnmtops | gs -q -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=- - > $output_file.pdf

To edit the default script for the scanner's Image option, I typed this:

cd /usr/local/Brother/sane/script
ls
sudo gedit scantoimage-0.2.1-3.sh

Once I was in the scantoimage script, I decided to change several things, including increasing the resolution to 300.  Now, unfortunately, the script didn’t work.  After I punched the Start button on the scanner, it just sat there for a moment, and then gave a long beep and a “sane_read: Error during device I/O,” and a tiny brscan.XXXXX file would appear in /home/ray; and when I opened that file, it was all black; or if I restored the line about echoing the output to gimp, it said this:

GIMP message
Opening ‘/home/ray/brscan/brscan.3bq3GA failed: PNM Image plug-in could not open image

and then gave up.  Weird thing is, I couldn’t get it to function normally, even when I restored scantoimage-0.2.1-3.sh to what I thought was its original condition.  I stopped and restarted brscan-skey (i.e., stop with brscan-skey -t); still the same thing.  I restarted the computer and, at the same time, shut down and then restarted the printer.  Having set brscan-skey to start automatically with Ubuntu, I went right to the console and tried again.  Now it scanned.  Maybe I had needed to completely kill brscan-skey after a bad edit.  I edited scantoimage-0.2.1-3.sh again, this time making just one change and then testing it.  Everything was fine until I got to the part where I made the recommended change to the actual scanimage line in the script.  Then, once again, the scanner just started the scan and then froze.  Eventually it reset itself.  I tried again, but now it was just waiting for a few seconds and then giving me that long beep and not even trying to scan.  I commented out their recommended change and restored the way it was originally, saved, and tried again.  Still just the long beep.  So I typed this

kill -9 `pidof brscan-skey-0.2.1-3`
brscan-skey

(note:  those are backticks, not single quotation marks) and then tried scanning again, but still the long beep and the error.  Turning the printer off for 30 seconds and then back on after killing brscan-skey did the trick.  I played some more and got the long beep and error again.  This time, I didn’t turn the printer off; I just killed brscan-skey and waited for the printer to reset itself.  That worked too.  After more searching, I found what looked like it might be an answer in Stutz’s Debian Linux Cookbook (pp. 286-287), which said this:

scanimage outputs images in the PNM (“portable anymap”) formats, so make sure that you have the `netpbm’ package (installed on most Linux systems by default); it’s a useful collection of tools for converting and manipulating these formats.

Well, Synaptic told me that I did not have netpbm installed.  I installed it, using Synaptic.  I observed that a list of devices (produced by “echo 'devicenames == ' | gs -q | tr " " "\n" | sort”) contained pnm but not the pnmtops device used in Brother’s command.  Pnmtops did appear to be in Ubuntu 10.04; it just didn’t appear to be on my system.  So I changed that too.  About this time, I discovered that Ubuntu had probably been giving me error messages in Terminal after each try, but Terminal tended to be buried under other windows.  So now I got this error:

scanimage: sane_read: Error during device I/O
scanimage: received signal 13
scanimage: trying to stop scannerSegmentation fault

So I changed it back to pnmtops.  This time, I got an additional error, after the “scanimage: sane_read: Error during device I/O” message:

pnmtops: warning, image too large for page, rescaling to 0.691703
pnmtops: writing color PostScript...
pnmtops: EOF/error reading 1 byte sample to file.

It looked like almost nobody had gotten that EOF/error message.  Not a good sign!  The rescaling part did not seem to be a problem.

Around this point, I realized that I had been trying to set the scanner’s Image option to produce PDFs, when I had originally wanted to do that via the File option.  I didn’t know whether I would have any better luck editing scantofile-0.2.1-3.sh than I had had with scantoimage-0.2.1-3.sh, but I decided to try.  I was puzzled that the Ubuntu Manual said, “Never use mktemp().”  I tried their recommended alternative of mkstemp, but bash did not seem to recognize it.  After some additional playing around, I came up with this working script:

#! /bin/sh

# $1 = scanner device
# $2 = friendly name
#
# Resolutions: 100,200,300,400,600

resolution=300
device=$1
sleep 0.01
output_file="/media/DATA/Current/Scan_`date +%Y%m%d-%H%M%S`.pdf"

scanimage --device-name "$device" --resolution $resolution | pnmtops | gs -q -dNOPAUSE -sDEVICE=pdfwrite -sOutputFile=- - > $output_file

chmod 644 $output_file

This script would name each PDF by its date and time, so that they would sort in the correct order in Nautilus or in Windows Explorer.  I think netpbm had to be installed (above) for this to work.  This script changed a number of things in the Brother script, including notably the filename and location.

I took these changes to the scantoimage script as well.  That script would normally output a PPM file, so I added a pnmtojpeg conversion line to produce an output file in the better-known JPG format, for maximum compatibility with various browsers and other programs.  I set quality to 95 because, in a brief test, the resulting JPG was half the size of one set to 100.  I included a slight bit of smoothing to make the file slightly smaller and better-looking, without blurring sharp lines.  I put this JPG file into a Lossy folder, to distinguish it from a parallel process in which I used the scanner’s PPM output to create a lossless PNG version, in case I needed maximum quality.  After the scanning session, I could then easily get rid of all unnecessary copies by deleting either the Lossy or the Lossless folder (or by mixing their contents as needed).  The resulting, working scantoimage script was as follows:

#! /bin/sh
#
# $1 = scanner device
# $2 = friendly name
#
# Resolution options: 100,200,300,400,600

resolution=300
device=$1
sleep 0.01
dir_name="/media/DATA/Current/"
file_name="Scan_`date +%Y%m%d-%H%M%S`"

ppm_file=$dir_name$file_name".ppm"
scanimage --device-name "$device" --resolution $resolution > $ppm_file

mkdir $dir_name"Lossy"
jpg_file=$dir_name"Lossy/"$file_name".jpg"
pnmtojpeg --quality=95 --smooth=10 $ppm_file > $jpg_file

mkdir $dir_name"Lossless"
png_file=$dir_name"Lossless/"$file_name".png"
pnmtopng $ppm_file > $png_file

# Change, comment, or delete the next line, depending on which
# file (if any) you want to open in GIMP
rm $ppm_file

# Uncomment the next line if you want the file to open in GIMP
# echo gimp $output_file \;rm -f $output_file | sh &

As with the printer setup, the final question was whether I could use the MFC-7340 from within a Windows XP virtual machine on VMware Workstation 7.  I went into a virtual machine, opened Adobe Acrobat, and entered the commands to scan.  It did not see any scanning devices.  I cancelled out of that and went to VM > Removable Devices > Brother Printer.  I saw that there was no “Brother Scanner” option, so this did not look good.  I clicked on Connect anyway, to connect the printer, and went back to Acrobat.  But that made the difference:  now it saw both TW-Brother MFC-7340 and WIA-Brother MFC-7340.  I usually used the latter, so I went with that.  I went through all the other normal steps to make a scan.  Unfortunately, Acrobat crashed.  It had been doing that anyway in that virtual machine, so I couldn’t infer anything for sure from that.  I tried again.  This time, Acrobat sat there for several minutes with the “Transferring data...” dialog open and then finally said, “Scanning canceled.”  I tried again, this time using the Twain (TW-Brother MFC-7340) option.  It said, “Reading from the device.”  It got as far as 0% Completed, and it hung there.  After maybe 10-15 minutes, I canceled it.  Apparently scanning would not be happening from within the virtual machine.  Otherwise, though, it appeared the project was complete.

How to Print a Long Webpage or Image File

I took this question to the Adobe Acrobat Windows forum. It drew a couple of responses, but no real answer. Here was the problem. Acrobat gives you the option of printing a webpage to PDF. Usually, it works just fine: you print the webpage, Acrobat breaks it up into a bunch of 8.5 x 11 sheets (if it's a long webpage), and you have a PDF document containing a reasonably good representation of the webpage. Sometimes, unfortunately, it does not work that way. Instead of printing the entire webpage to PDF, Acrobat prints just the first and last pages, or maybe just the first page. I think the reason must have to do with the HTML coding of the webpage. Whatever: point is, you can't PDF the webpage. This happens for some long image files too. For instance, I thought of using the ScreenGrab extension in Firefox to save the irritating long webpage to JPG or PNG format, and then using an image editor (e.g., the highly recommended freeware IrfanView) to print the PNG to PDF. But this didn't work either: I still got the same outcome. Likewise if I first saved the webpage to different forms of HTML files on my local drive. Eventually, though, I came to a simple solution. Instead of trying to print to 8.5 x 11-inch paper, save the long webpage to a PNG, and then set Acrobat to print to a sheet that is 92 x 92 inches. There are many webpages that are still too long for that, but it's not a bad size. If you need to convert that outcome to 8.5 x 11, then maybe you can print the 92" PDF to an 8.5 x 11 PDF size. In my experiments so far, this approach gives me fonts that look pretty much normal, viewed at a page width display setting. They are good enough to OCR in Acrobat.