DMPEG V1.1

Public Domain MPEG decoder

by Stefan Eckart

June 1993

1. Features
===========

DMPEG is another MPEG decoder/player for the PC:


- decodes (nearly) the full MPEG video standard
(I,P,B frames, frame size up to at least 352x288)

- can save decoded sequence in 8 or 24bit raw file for
fast off-line display (two pass mode)

- optional on-screen display during decoding

- several dithering options for 8 bit displays:
ordered dither, Floyd-Steinberg, grayscale

- selectable color-space

- runs under DOS, 640KB RAM, no MS-Windows or '386 required

- compact (small code / small data models, 16 bit arithmetic)

- supports VGA, many Super-VGAs (including VESA) and
some TrueColor SVGAs


2. Introduction
===============

DMPEG is a Public Domain decoder for files containing MPEG video
sequences. MPEG is a standardized format for compressed storage of video
(and audio), similar in concept to JPEG for individual images. It,
however, provides higher compression by exploiting the similarity of
consecutive images of a sequence, so called interframe coding. As with
JPEG, the compression factor depends on the source material and the
desired quality of the coded sequence. A compression to less than 5% of
the original material is not uncommon.

DMPEG supports the full MPEG video standard, except D-frame sequences.
Decoding is not very fast, therefore a two-pass mode was implemented:
during the first pass the sequence is decoded and stored in an unencoded
file which then can be played at maximum speed during the second pass.
Of course DMPEG also has a one-pass mode where the sequence is displayed
immediately but at lower speed. The speed is about one seventh of that
of the Xing MPEG decoder (see appendix A). The Xing player, however, can
only decode I frame sequences of size 160x120 and needs a '386. I
suggest to use DMEG in addition to Xing's MPEG.EXE 320x200 DOS player.
The latter is appropriate for real-time display of '160x120, I-frame
only' Xing files, DMPEG is more suitable for larger size 'real' MPEGs
containing I, P and B type, 352x288 sized frames or for faster and
better quality 'off-line' display of the 160x120 MPEGs. Try it on one of
the larger MPEG files (see appendix, my recommendation: flowg.mpg) to
see that MPEG can also be used (and is intended) for storing really
sharp, virtually noise-free sequences.

To give an impression of the attainable speeds, here are some figures
measured on a 386DX/33 with 4MB RAM and a Conner 3104 100 MB hard disk.
The decoding time for flowg.mpg (352x240) was about 4 sec per frame, 9
minutes for 150 frames. waterski.mpg (336x208 and higher compression)
required only 2.4 sec per frame. A decoded 150 frame raw data file is
12 MB long and can be displayed at 5.2 frames/s. 160x120 MPEGs reach
22 frames/s while the decoding speed is 1.9 sec/frame. The display frame
rate is limited by the transfer speed from hard disk to memory (about
550 kByte/s on my PC). If you have enough RAM, you can obtain much
higher speed by playing from a large RAM-disk (e.g 23 frames/s for a
352x240 sequence).

This program is Public Domain and I don't take any responsibility
regarding its fitness, usefulness etc. (#include
). Comments, bug reports, questions to:

Stefan Eckart
Kagerstr. 4
D-81669 Muenchen
Germany

email: [email protected]

Any feedback is welcome.


3. Usage
========

dmpeg [options] input.mpg [output.raw]

dmpeg [options] input.raw

input.mpg any MPEG DIS 11172 compliant non multiplexed compression
layer video stream (D frame streams unsupported)

output.raw the decoded and dithered 8 or 24bit raw output file;
if omitted: output to screen only

input.raw previously decoded raw file to be displayed at maximum speed

Options:

-q quiet mode; no text output (except error messages)
-v verbosity level; can be increased by repeating this option;
enables display of decoded information (start codes, frame size,
quantization scales etc.), can't be combined with the -s option

-dx dithering options:

-d0 ordered, saturation dominant 4x4 dither (default)
-d1 Floyd-Steinberg error-diffusion / blue-noise-shaping (2 weights)
-d2 Floyd-Steinberg error-diffusion / blue-noise-shaping (4 weights)
-d3 undithered grayscale output
-d4 24 bit true color output

-p0 use full color range palette
-p1 use restricted color range palette (default)

-sn select graphics adapter:

-s0 either standard VGA or Super-VGA in 320x200x256 mode (default)
(options -d0..3) or a TrueColor VGA with VESA BIOS in
320x200x16M mode (option -d4)
-s1 VESA BIOS, either 640x480x256 (-d0..3) or 640x480x16M (-d4)
-s2 ET4000 \
-s3 ET3000 |
-s4 Video 7 |
-s5 Paradise | 640x480x256
-s6 Trident | (options -d0..3 only)
-s7 Chips & Technologies |
-s8 ATI |
-s9 Orchid |
-s10 Oaktech /

-lx delay (x=0..65535, default 0), used only for offline display

-b use 8 bit transfer (default: 16 bit) into (S)VGA memory


Options can be combined. Example:

dmpeg -d2s0 flowg.mpg flowg.raw

decodes MPEG file flowg.mpg to raw file flowg.raw using FS4 dithering
and displays it on the screen in 320x200x256 mode.

dmpeg -vvv flowg.mpg flowg.raw

as above but using ordered dither and printing much information about
startcodes, headers and parameters.

All SVGA drivers use the 640x480x256 display mode and require at least
512 kB of video memory. If your graphics card is not listed, the best
you can do (besides trying all drivers in the hope that one of them
might work) is to obtain a VESA BIOS extension for your card and use
mode -s1. A collection of such drivers had been posted to
comp.binaries.ibm.pc and is also available at
ftp.rahul.net:/pub/bryanw/pc/vesadrv2.zip (anonymous ftp).

All drivers except the standard VGA, ET4000, the VESA driver in
conjunction with the mentioned public domain VESA BIOS extension TSR for
the ET4000 and the TrueColor driver with a Cirrus Logic VGA are
untested. I'm very interested in email feedback which drivers work and
which don't.

The -l option controls playback speed. It is implemented as a simple
delay loop without synchronization to a timer or vertical retrace.

The -b option is probably superfluous. I don't have much experience in
PC graphics programming and since I saw some example driver routines
using bytewise transfer (rep movsb) instead of wordwise (rep movsw) I
preferred to include this as a fudge factor. You should try it only in
case of problems. Speed is reduced considerably if you activate this
switch (at least for cards with 16 bit bus interface).

Decoding can be terminated by pressing any key.

Offline display can be controlled with the following keys:

any key except
space, return
or escape halt display at current frame and step one frame
forward each time a key is pressed

space step one frame backwards

return continue display

escape quit program


4. Technical information
========================

The player is a rather straightforward implementation of the MPEG spec
[1]. The IDCT is based on the Chen-Wang 11 multiplication algorithm [2]
which was coded in assembler but does not use tables for multiplication.
Blocks with not more than two non-zero coefficients use a non-separated
direct multiply-accumulate 2D-IDCT, which turned out to be faster than a
'fast' algorithm in this (quite common) case. Dithering is pretty
standard. Main difference to the Berkeley decoder (except for the fewer
number of supported algorithms) is the use of 256 instead of 128 colors,
the (default) option to use a restricted color-space and the
implementation of a color saturation dominant ordered dither. This leads
to a somewhat superior quality of the dithered image.

Restricted color-space means that the U and V components are clipped to
+/-0.25 (instead of +/-0.5) and the display color-space points are
distributed over this restricted space. Since the distance between
color-space points is thus reduced by a factor of two, the color
resolution is doubled at the expense of not being able to represent
fully saturated colors.

Saturation dominant ordered dither is a method by which a color, lying
somewhere between the points of the display color-space, is approximated
by primarily alternating between two points of constant hue instead of
constant saturation. This yields subjectivly better quality due to the
lower sensitivity of the human viewing system to saturation changes than
to hue changes (the same reasoning as used by the PAL TV standard to
improve on NTSC). The improvement is particularly visible in dark brown
or redish areas.


5. File formats
===============

If you want to write your own player or to post-process the results,
here is the format of the 8 bit raw file:

Byte

0..7 compatibility bytes (to be ignored)
8..9 image width (MSB first)
10..11 image height (MSB first)
12..31 compatibility / unused
32..799 color table, R[0],G[0],B[0], ... R[255],G[255],B[255]
800.. image data in natural order (top left to bottom right,
all frames concatenated without any gaps or repeated
headers)

The format of 24bit true color files (-d4 option) is:

0..7 compatibility bytes (to be ignored)
8..9 image width (MSB first)
10..11 image height (MSB first)
12..31 compatibility / unused
32.. image data (3 bytes per pixel: B,G,R, top left to bottom
right, all frames concatenated without any gaps or repeated
headers)

The 8 bit format happens to be compatible with the raw format used by a
shareware program called Image Alchemy. In fact you can use that program
to view the first frame of the sequence and to convert it into other
formats.


6. Changes
==========

This is a list of major changes relative to version 1.0.

- speed improved by a factor of two through recoding of the IDCT and
interframe prediction routines in assembler and optimization of the
dithering algorithms

- support of TrueColor VGAs with VESA BIOS (only tested with Cirrus
Logic)

- DMPEG and DMPLAY merged into one program

- SVGA / TrueColor support for immediate display during decoding

- increased robustness against errors in the file to be decoded:
program tries to resynchronize on the next slice/frame header
instead of terminating

- format of the 24 bit raw file changed (B,G,R instead of R,G,B)


7. References
=============

1. Coding of moving pictures and associated audio for digital storage
media up to about 1,5 Mbit/s, Draft International Standard ISO/IEC
DIS 11172, 1992.

2. Chen, Wang, IEEE ASSP-32, pp. 803-816, Aug. 1984.


Appendix A: Related Software
============================

This list is probably incomplete, but it's all I'm aware of. Of course
there are programs for other systems as well (Mac, Amiga etc.).

cmpeg an MPEG encoder for the PC (DOS, 640K, no '386 req.)
for Targa, PBMPLUS and Alchemy RAW images
Author: myself

mpeg_play MPEG Video Software Decoder (Version 2.0; Jan 27, 1993)
Authors: Lawrence A. Rowe, Ketan Patel, and Brian Smith
Computer Science Division-EECS, Univ. of Calif. at
Berkeley
toe.cs.berkeley.edu:/pub/multimedia/mpeg/mpeg-2.0.tar.Z

mpgplay Online port of mpeg_play for DOS
by: Giampero Caprino, [email protected]
(evaluation version, works only with Xing files)
mpeg386.exe Offline port of mpeg_play for DOS
by: Greg Ennis, [email protected]
(based on mpgplay, works only with Xing files)
mpegwin Online port of mpeg_play for MS-Windows
by: Michael Simmons, [email protected]
toe.cs.berkeley.edu:/pub/multimedia/mpeg/Ports/mpegw*
(HiColor & TrueColor support, Shareware)

mplay.exe,
mpeg.exe DOS MPEG players from Xing Technologies
(very high speed, but decodes only a small subset of the
MPEG standard)

MPEGv1.1/1.2alpha
MPEG Software Encoder/Decoder
Authors: Portable Video Research Group (PVRG)
havefun.stanford.edu:/pub/mpeg/MPEGv*.tar.Z

mpgcodec PVRG encoder/decoder for PC compiled with GNU gcc
('386 required)
posted in alt.binaries.pictures.utilities


APPENDIX B: MPEG files
======================

Two good sources for MPEG files:

toe.cs.berkeley.edu:/pub/multimedia/mpeg/movies
havefun.stanford.edu:/pub/mpeg

High quality MPEGs you simply can't afford to miss:

tennis.mpg
flowg.mpg
bike.mpg

--
Stefan Eckart, [email protected], June 1993.