Netpbm is a package of graphics programs and a programming library.
There are over 220 separate programs in the package, most of which have "pbm", "pgm", "ppm", "pam", or "pnm" in their names. For example, pamscale and giftopnm.
For example, you might use pamscale to shrink an image by 10%. Or use pamcomp to overlay one image on top of another. Or use pbmtext to create an image of text. Or reduce the number of colors in an image with pnmquant.
Netpbm is an open source software package, distributed via the Sourceforge netpbm project.
The Netpbm programs are generally useful run by a person from a command shell, but are also designed to be used by programs. A common characteristic of Netpbm programs is that they are simple, fundamental building blocks. They are most powerful when stacked in pipelines. Netpbm programs do not use graphical user interfaces and do not seek input from a user. The only programs that display graphics at all are the very primitive display programs pamx and ppmsvgalib, and they don't do anything but that.
Each of these programs has its own manual, as linked in the directory below.
The Netpbm programs can read and write files greater than 2 GiB wherever the underlying system can. There may be exceptions where the programs use external libraries (The JPEG library, etc.) to access files and the external library does not have large file capability. Before Netpbm 10.15 (April 2003), no Netpbm program could read a file that large.
There are a few options that are present on all programs that are based on the Netpbm library, including virtually all Netpbm programs. These are not mentioned in the individual manuals for the programs.
You can use two hyphens instead of one on these options if you like.
This option was introduced in Netpbm 10.10 (October 2002).
Here is a complete list of all the Netpbm programs (with links to their manuals):
As a collection of primitive tools, the power of Netpbm is multiplied by the power of all the other unix tools you can use with them. These notes remind you of some of the more useful ways to do this. Often, when people want to add high level functions to the Netpbm tools, they have overlooked some existing tool that, in combination with Netpbm, already does it.
Often, you need to apply some conversion or edit to a whole bunch of files.
As a rule, Netpbm programs take one input file and produce one output file, usually on Standard Output. This is for flexibility, since you so often have to pipeline many tools together.
Here is an example of a shell command to convert all your of PNG files (named *.png) to JPEG files named *.jpg:
for i in *.png; do pngtopam $i | ppmtojpeg >`basename $i .png`.jpg; done
Or you might just generate a stream of individual shell commands, one per file, with awk or perl. Here's how to brighten 30 YUV images that make up one second of a movie, keeping the images in the same files:
ls *.yuv | perl -ne 'chomp; print yuvtoppm $_ | ppmbrighten -v 100 | ppmtoyuv >tmp$$.yuv; mv tmp$$.yuv $_ ' | sh
The tools find (with the -exec option) and xargs are also useful for simple manipulation of groups of files.
Some shells' "process substitution" facility can help where a non-Netpbm program expects you to identify a disk file for input and you want it to use the result of a Netpbm manipulation. Say the hypothetical program printcmyk takes the filename of a Tiff CMYK file as input and what you have is a PNG file abc.png. Try:
printcmyk <({ pngtopam abc.png | pnmtotiffcmyk ; })
It works in the other direction too, if you have a program that makes you name its output file and you want the output to go through a Netpbm tool.
All of the programs work with a set of graphics formats called the "netpbm" formats. Specifically, these formats are pbm, pgm, ppm, and pam. The first three of these are sometimes known generically as "pnm". Many of the Netpbm programs convert from a Netpbm format to another format or vice versa. This is so you can use the Netpbm programs to work on graphics of any format. It is also common to use a combination of Netpbm programs to convert from one non-Netpbm format to another non-Netpbm format. Netpbm has converters for about 100 graphics formats, and as a package Netpbm lets you do more graphics format conversions than any other computer graphics facility.
The Netpbm formats are all raster formats, i.e. they describe an image as a matrix of rows and columns of pixels. In the PBM format, the pixels are black and white. In the PGM format, pixels are shades of gray. In the PPM format, the pixels are in full color. The PAM format is more sophisticated. A replacement for all three of the other formats, it can represent matrices of general data including but not limited to black and white, grayscale, and color images.
Programs designed to work with PBM images have "pbm" in their names. Programs designed to work with PGM, PPM, and PAM images similarly have "pgm", "ppm", and "pam" in their names.
All Netpbm programs designed to read PGM images see PBM images as if they were PGM too. All Netpbm programs designed to read PPM images see PGM and PBM images as if they were PPM. See Implied Format Conversion.
Programs that have "pnm" in their names read PBM, PGM, and PPM but unlike "ppm" programs, they distinguish between those formats and their function depends on the format. For example, pnmtopng creates a black and white PNG output image if its input is PBM or PGM, but a color PNG output image if its input is PPM. And pnmrotate produces an output image of the same format as the input. A hypothetical ppmrotate program would also read all three PNM input formats, but would see them all as PPM and would always generate PPM output.
Programs that have "pam" in their names read all the Netpbm formats: PBM, PGM, PPM, and PAM. They sometimes treat them all as if they are PAM, using an implied conversion, but often they recognize the individual formats and behave accordingly, like a "pnm" program does. See Implied Format Conversion.
Finally, there are subformats of PAM that are equivalent to PBM, PGM, and PPM respectively, and Netpbm programs designed to read PBM, PGM, and/or PPM see those PAM images as if they were the former. For example, ppmhist can analyze a PAM image of tuple type RGB (i.e. a color image) as if it were PPM.
If it seems wasteful to you to have three separate PNM formats, be aware that there is a historical reason for it. In the beginning, there were only PBMs. PGMs came later, and then PPMs. Much later came PAM, which realizes the possibility of having just one aggregate format.
The formats are described in the specifications of pbm, pgm, ppm, and pam.
A program that uses the PGM library subroutines to read an image can read a PBM image as well as a PGM image. The program sees the PBM image as if it were the equivalent PGM image, with a maxval of 255. note: This sometimes confuses people who are looking at the formats at a lower layer than they ought to be because a zero value in a PBM raster means white, while a zero value in a PGM raster means black.
A program that uses the PPM library subroutines to read an image can read a PGM image as well as a PPM image and a PBM image as well as a PGM image. The program sees the PBM or PGM image as if it were the equivalent PPM image, with a maxval of 255 in the PBM case and the same maxval as the PGM in the PGM case.
A program that uses the PAM library subroutines to read an image can read a PBM, PGM, or PPM image as well as a PAM image. The program sees a PBM image as if it were the equivalent PAM image with tuple type BLACKANDWHITE. It sees a PGM image as if it were the equivalent PAM image with tuple type GRAYSCALE. It sees a PPM image as if it were the equivalent PAM image with tuple type RGB. But the program actually can see deeper if it wants to. It can tell exactly which format the input was and may respond accordingly. For example, a PAM program typically produces output in the same format as its input.
A program that uses the PGM library subroutines to read an image can read a PAM image as well a PGM image, if the PAM is a grayscale or black and white visual image. That canonically means the PAM has a depth of 1 and a tuple type of GRAYSCALE or BLACKANDWHITE, but most Netpbm programs are fairly liberal and will take any PAM at all, ignoring all but the first plane.
There is a similar implied conversion for PPM library subroutines reading PAM. There is nothing similar for PBM, so if you need for a PBM program to read a PAM image, run it through pamtopnm.
In many graphics formats, there's a means of indicating that certain parts of the image are wholly or partially transparent, meaning that if it were displayed "over" another image, the other image would show through there. Netpbm formats deliberately omit that capability, since their purpose is to be extremely simple.
In Netpbm, you handle transparency via a transparency mask in a separate (slightly redefined) PGM image. In this pseudo-PGM, what would normally be a pixel's intensity is instead an opaqueness value. See pgm. pamcomp is an example of a program that uses a PGM transparency mask.
Another means of representing transparency information has recently developed in Netpbm, using PAM images. In spite of the argument given above that Netpbm formats should be too simple to have transparency information built in, it turns out to be extremely inconvenient to have to carry the transparency information around separately. This is primarily because Unix shells don't provide easy ways to have networks of pipelines. You get one input and one output from each program in a pipeline. So you'd like to have both the color information and the transparency information for an image in the same pipe at the same time.
For that reason, some new (and recently renovated) Netpbm programs recognize and generate a PAM image with tuple type RGB_ALPHA or GRAYSCALE_ALPHA, which contains a plane for the transparency information. See the PAM specification.
The Netpbm programming library, libnetpbm, makes it easy to write programs that manipulate graphic images. Its main function is to read and write files in the Netpbm formats, and because the Netpbm package contains converters for all the popular graphics formats, if your program reads and writes the Netpbm formats, you can use it with any formats.
But the library also contain some utility functions, such as character drawing and RGB/YCrCb conversion.
The library has the conventional C linkage. Virtually all programs in the Netpbm package are based on the Netpbm library.
In a standard installation of Netpbm, there is a program named netpbm-config in the regular program search path. We don't consider this a Netpbm program -- it's just an ancillary part of a Netpbm installation. This program tells you information about the Netpbm installation, and is intended to be run by other programs that interface with Netpbm. In fact, netpbm-config is really a configuration file, like those you typically see in the /etc/ directory of a Unix system.
Example:
$netpbm-config --datadir
/usr/local/netpbm/data
If you write a program that needs to access a Netpbm data file, it can
use such a shell command to find out where the Netpbm data files are.
netpbm-config is the only file that must be installed in a standard directory (it must be in a directory that is in the default program search path). You can use netpbm-config as a bootstrap to find all the other Netpbm files.
There is no detailed documentation of netpbm-config. If you're in a position to use it, you should have no trouble reading the file itself to figure out how to use it.
An important characteristic that varies among graphics software is how much memory it uses, and how. Does it read an entire image into memory, work on it there, then write it out all at once? Does it read one and write one pixel at a time? In Netpbm, it differs from one program to the next, but there are some generalizations we can make.
Most Netpbm programs keep one row of pixels at a time in memory. Such a program reads a row from an input file, processes it, then writes a row to an output file. Some programs execute algorithms that can't work like that, so they keep a small window of rows in memory. Others must keep the entire image in memory. If you think of what job the program does, you can probably guess which one it does.
When Netpbm keeps a pixel in memory, it normally uses a lot more space for it than it occupies in the Netpbm image file format.
The older programs (most of Netpbm) use 12 bytes per pixel. This is true even for a PBM image, for which it only really takes one bit to totally describe the pixel. Netpbm does this expansion to make implementing the programs easier -- it uses the same format regardless of the type of image.
Newer programs use the "pam" family of library functions internally, which use memory a little differently. These functions are designed to handle generic tuples with a variable numbers of planes, so no fixed size per-tuple storage is possible. A program of this type uses 4 bytes per sample (a tuple is composed of samples), plus a pointer (4-8 bytes) per tuple. In a graphic image, a tuple is a pixel. So an ordinary color image takes 16-20 bytes per pixel.
When considering memory usage, it is important to remember that memory and disk storage are equivalent in two ways:
This means that the consequences of whether a program works from the image file or from a memory copy are not straightforward.
Note that an image takes a lot less space in a Netpbm format file, and therefore in an operating system's file cache, than in Netpbm's in-memory format. In non-Netpbm image formats, the data is even smaller. So reading through an input file multiple times instead of keeping a copy in regular memory can be the best use of memory, and many Netpbm programs do that. But some files can't be read multiple times. In particular, you can't rewind and re-read a pipe, and a pipe is often the input for a Netpbm program. Netpbm programs that re-read files detect such input files and read them into a temporary file, then read that temporary file multiple times.
A few Netpbm programs use an in-memory format that is just one bit per pixel. These are programs that convert between PBM and a format that has a raster format very much like PBM's. In this case, it would actually make the program more complicated (in addition to much slower) to use Netpbm's generic 12 byte or 8 byte pixel representation.
By the way, the old axiom that memory is way faster than disk is not necessarily true. On small systems, it typically is true, but on a system with a large network of disks, especially with striping, it is quite easy for the disk storage to be capable of supplying data faster than the CPU can use it.
People sometimes wonder what CPU facilities Netpbm programs and the Netpbm programming library use. The programs never depend on particular features existing (assuming they're compiled properly), but the speed and cost of running a program varies depending upon the CPU features.
Note that when you download a binary that someone else compiled, even though it appears to be compiled properly for your machine, it may be compiled improperly for that machine if it is old, because the person who compiled it may have chosen to exploit features of newer CPUs in the line. For example, an x86 program may be compiled to use instructions that are present on an 80486, but not on an 80386. You would probably not know this until you run the program and it crashes.
But the default build options almost always build binaries that are as backward compatible with old CPUs as possible. An exception is a build for a 64 bit x86 CPU. While the builder could build a program that runs on a 32 bit x86, it does not do so by default. A default build builds a program will not run on an older 32-bit-only x86 CPU.
One common build option is to use MMX/SSE operands with x86 CPUs. Those are not available on older x86 CPUs. The builder by default does not generate code that uses MMX/SSE when building for 32 bit x86 CPUs, but does when building for 64 bit x86.
One area of particular importance is floating point arithmetic. The Netpbm image formats are based on integers, and Netpbm arithmetic is done with integers where possible. But there is one significant area that is floating point: programs that must deal with light intensity. The Netpbm formats use integers that are proportional to brightness, and brightness is exponentially related to light intensity. The programs have to keep the intermediate intensity values in floating point in order not to lose precision. And the conversion (gamma function) between the two is heavy-duty floating point arithmetic. Programs that mix pixels together have to combine light intensity, so they do heavy floating point. Three of the most popular Netpbm programs do that: pamscale (shrink/expand an image), pamcomp (overlay an image over another one), and pamditherbw (Make a black and white image that approximates a grayscale image).
The Netpbm image formats use 16 bit integers. The Netpbm code uses "unsigned int" size integers to work with them.
The Gimp is a visual image editor for Unix and X, so it does the kinds of things that Netpbm does, but interactively in a user-friendly way. The Gimp knows a variety of graphics file formats and image transformations, but you can extend it with plugins.
A particularly easy way to write a Gimp plugin is to write a Netpbm program (remember that a fundamental mission of Netpbm is make writing image manipulation programs easy) and then use netpbm2gimp to compile that same source code into a Gimp plugin.
You can turn a program that converts from a certain graphics file format to Netpbm format into a Gimp load plugin. Likewise, you can turn a program that converts to a certain graphics format from Netpbm format into a Gimp store plugin. Finally, a program that transforms images in Netpbm format can become a process plugin.
And the netpbm2gimp project has already packaged for you a few hundred of the Netpbm programs as Gimp plugins. With this package you can, for example, edit an image in any of the arcane graphics file formats that Netpbm understands but no other image editor in existence does.
If you're using Netpbm to do graphics for a website, you can invoke the Netpbm programs from a PHP script. To make this even easier, check out PHP-NetPBM, a PHP class that interacts with Netpbm. Its main goal is to decrease the pain of using Netpbm when working with images in various formats. It includes macro commands to perform manipulations on many files.
I can't actually recommend PHP-NetPBM. I spent some time staring at it and was unable to make sense of it. Some documentation is in fractured English and other is in an unusual character set. But a PHP expert might be able to figure it out and get some use out of it.
Netpbm contains primitive building blocks. It certainly is not a complete graphics software library.
The first thing you will want to make use of any of these tools is a viewer. (On GNU/Linux, you can use Netpbm's pamx or ppmsvgalib in a pinch, but it is pretty limiting). zgv is a good full service viewer to use on a GNU/Linux system with the SVGALIB graphics display driver library. You can find zgv at ftp://ftp.ibiblio.org/pub/Linux/apps/graphics/viewers/svga.
zgv even has a feature in it wherein you can visually crop an image and write an output file of the cropped image using pamcut. See the -s option to zgv.
For the X inclined, there is also xzgv.
xwud (X Window Undump) is a classic application program in the X Window System that displays an image in an X window. It takes the special X Window Dump format as input; you can use Netpbm's pnmtoxwd to create it. You're probably better off just using Netpbm's pamx.
xloadimage and its extension xli are also common ways to display a graphic image in X.
gqview is a more modern X-based image viewer.
qiv is a small, very fast viewer for X.
To play mpeg movies, such as produced by ppmtompeg, try mplayer or xine.
See ftp://metalab.unc.edu/pub/Linux/apps/graphics/viewers/X.
xwd (X Window Dump), a classic application program in the X Window System, captures the contents of an X window, in its own special image format, called X Window Dump File. You can use Netpbm's xwdtopnm to turn it into something more useful.
fbdump Capturers the current contents of a video display on the local computer and generates a PPM image of it. It works with Linux framebuffer devices.
Visual graphics software is modern point-and-click software that displays an image and lets you work on it and see the results as you go. This is fundamentally different from what Netpbm programs do.
ImageMagick is like a visual version of Netpbm. Using the X/Window system on Unix, you can do basic editing of images and lots of format conversions. The package does include at least some non-visual tools. convert, mogrify, montage, and animate are popular programs from the ImageMagick package. ImageMagick runs on Unix, Windows, Windows NT, Macintosh, and VMS.
xv is a very old and very popular simple image editor in the Unix world. It does not have much in the way of current support, or maintenance, though.
The Gimp is a visual image editor for Unix and X, in the same category as the more famous, less capable, and much more expensive Adobe Photoshop, etc. for Windows. See http://www.gimp.org. And you can add most of Netpbm's function to The Gimp using Netpbm2gimp.
Electric Eyes, kuickshow, and gthumb are also visual editors for the X/Window system, and KView and gwenview are specifically for KDE.
If you're writing a program in C to draw and manipulate images, check out gd. Netpbm contains a C library for drawing images (libnetpbm's "ppmd" routines), but it is probably not as capable or documented as gd. You can easily run any Netpbm program from a C program with the pm_system function from the Netpbm programming library, but that is less efficient than gd functions that do the same thing.
Cairo is similar.
Ilib is a C subroutine library with functions for adding text to an image (as you might do at a higher level with pbmtext, pamcomp, etc.). It works with Netpbm input and output. Find it at k5n.us. Netpbm also includes character drawing functions in the libnetpbm library, but they do not have as fancy font capabilities (see ppmdraw for an example of use of the Netpbm character drawing functions).
Pango is another text rendering library, with an emphasis on internationalization.
Pango and Cairo complement each other and work well together.
GD is a library of graphics routines that is part of PHP. It has a subset of Netpbm's functions and has been found to resize images more slowly and with less quality.
mencode, which is part of the mplayer package, creates movie files. It's like a much more advanced version of ppmtompeg, without the Netpbm building block simplicity.
MJPEGTools is software for dealing with the MJPEG movie format.
To create an animated GIF, or extract a frame from one, use gifsicle. gifsicle converts between animated GIF and still GIF, and you can use pamtogif and giftopnm to connect up to all the Netpbm utilities. See http://www.lcdf.org/gifsicle.
To convert an image of text to text (optical character recognition - OCR), use gocr (think of it as an inverse of pbmtext). See http://jocr.sourceforge.net/.
http://schaik.com/pngsuite contains a PNG test suite -- a whole bunch of PNG images exploiting the various features of the PNG format.
Other versions of Netpbm's pnmtopng/pngtopam are at http://www.schaik.com/png/pnmtopng.html.
The version in Netpbm was actually based on that package a long time ago, and you can expect to find better exploitation of the PNG format, especially recent enhancements, in that package. It may be a little less consistent with the Netpbm project and less exploitive of recent Netpbm format enhancements, though.
pngwriter is a C++ library for creating PNG images. With it, you plot an image pixel by pixel. You can also render text with the FreeType2 library.
jpegtran Does some of the same transformations as Netpbm is famous for, but does them specifically on JPEG files and does them without loss of information. By contrast, if you were to use Netpbm, you would first decompress the JPEG image to Netpbm format, then transform the image, then compress it back to JPEG format. In that recompression, you lose a little image information because JPEG is a lossy compression. Of course, only a few kinds of lossless transformation are possible. jpegtran comes with the Independent JPEG Group's (http://www.ijg.org) JPEG library.
Some tools to deal with EXIF files (see also Netpbm's jpegtopnm and pnmtojpeg): To dump (interpret) an EXIF header: Exifdump ((http://topo.math.u-psud.fr/~bousch/exifdump.py)) or Jhead.
A Python EXIF library and dumper: http://pyexif.sourceforge.net.
Here's some software to work with IOCA (Image Object Content Architecture): ImageToolbox ($2500, demo available). This can convert from TIFF -> IOCA and back again. Ameri-Imager ($40 Windows only).
pnm2ppa converts to HP's "Winprinter" format (for HP 710, 720, 820, 1000, etc). It is a superset of Netpbm's pbmtoppa and handles, notably, color. However, it is more of a printer driver than a Netpbm-style primitive graphics building block. See The Pnm2ppa /Sourceforge Project
DjVuLibre is a package of software for using the DjVu format. It includes viewers, browser plugins, decoders, simple encoders, and utilities. The encoders and decoders can convert between DjVu and PNM. See the DjVu website.
There is a large class of software that does document processing, and that is somewhat related to graphics because documents contain graphics and a page of a document is for many purposes a graphic image. Because of this slight intersection with graphics, I cover document processing software here briefly, but it is for the most part beyond the scope of this document.
First, we look at where Netpbm meets document processing. pstopnm converts from Postscript and PDF to PNM. It effectively renders the document into images of printed pages. pstopnm is nothing but a convenient wrapper for Ghostscript, and in particular Netpbm-format device drivers that are part of it. pnmtops and pbmtoepsi convert a PNM image to a Postscript program for printing the image. But to really use PDF and Postscript files, you generally need more complex document processing software.
Adobe invented Postscript and PDF and products from Adobe are for many purposes the quintessential Postscript and PDF tools.
Adobe's free Acrobat Reader displays PDF and converts to Postscript. The Acrobat Reader for unix has a program name of "acroread" and the -toPostScript option (also see the -level2 option) is useful.
Other software from Adobe, available for purchase, interprets and creates Postscript and PDF files. "Distill" is a program that converts Postscript to PDF.
xpdf also reads PDF files.
GSview, ghostview, gv, ggv, and kghostview are some other viewers for Postscript and PDF files.
The program ps2pdf, part of Ghostscript, converts from Postscript to PDF.
Two packages that produce more kinds of Encapsulated Postscript than the Netpbm programs, including compressed kinds, are bmeps and imgtops.
dvips converts from DVI format to Postscript. DVI is the format that Tex produces. Netpbm can convert from Postscript to PNM. Thus, you can use these in combination to work with Tex/Latex documents graphically.
wvware converts a Microsoft Word document (.doc file) to various other formats. While the web page doesn't seem to mention it, it reportedly can extract an embedded image in a Word document as a PNG.
Document Printer converts various print document formats (Microsoft Word, PDF, HTML, etc.) to various graphic image formats. ($38, Windows only).
Latex2html converts Latex document source to HTML document source. Part of that involves graphics, and Latex2html uses Netpbm tools for some of that. But Latex2html through its history has had some rather esoteric codependencies with Netpbm. Older Latex2html doesn't work with current Netpbm. Latex2html-99.2beta8 works, though.
The file program looks at a file and tells you what kind of file it is. It recognizes most of the graphics formats with which Netpbm deals, so it is pretty handy for graphics work. Netpbm's anytopnm program depends on file. See ftp://ftp.astron.com/pub/file.
The Utah Raster Toolkit serves a lot of the same purpose as Netpbm, but without the emphasis on format conversions. This package is based on the RLE format, which you can convert to and from the Netpbm formats. The website of the Geometric Design And Computation group in the Department of Computer Science at University of Utah used to (ca. 2002) have information on the Utah Raster Toolkit, but now it doesn't.
Ivtools is a suite of free X Window System drawing editors for Postscript, Tex, and web graphics production, as well as an embeddable and extendable vector graphic shell. It uses the Netpbm facilities. See http://www.ivtools.org.
Chisato Yamauchi <cyamauch@ir.isas.jaxa.jp> has written a free c/Fortran graphic library: EGGX/ProCall. He says he tried to write the ultimate easy-to-use graphic kit for X. It is for drawing upon an X11 window, but for storage, it outputs PPM. He suggests Netpbm to convert to other formats.
The program morph morphs one image into another. It uses Targa format images, but you can use tgatoppm and ppmtotga to deal with that format. You have to use the graphical (X/Tk) Xmorph to create the mesh files that you must feed to morph. morph is part of the Xmorph package. See http://xmorph.sourceforge.net/.
People never seem to tire of inventing new graphics formats, often completely redundant with pre-existing ones. Netpbm cannot keep up with them. Here is a list of a few that we know Netpbm does not handle (yet).
Various commercial Windows software handles dozens of formats that Netpbm does not, especially formats typically used with Windows programs. ImageMagick is probably the most used free image format converter and it also handles lots of formats Netpbm does not.
For more information, see the DjVu website.
The web page http://www.sollers.ca describes a program for converting from CALS to TIFF.
Netpbm has a long history, starting with Jef Poskanzer's Pbmplus package in 1988. See the Netpbm web site for details.
The file doc/HISTORY in the Netpbm source code contains a detailed change history release by release.
Netpbm is based on the Pbmplus package by Jef Poskanzer, first distributed in 1988 and maintained by him until 1991. But the package contains work by countless other authors, added since Jef's original work. In fact, the name is derived from the fact that the work was contributed by people all over the world via the Internet, when such collaboration was still novel enough to merit naming the package after it.
Bryan Henderson has been maintaining Netpbm since 1999. In addition to packaging work by others, Bryan has also written a significant amount of new material for the package.