Libris Britannia 4/science library(b).zip/science library(b)/INFO/PCCDEMO.ZIP/COMP1.EXE/JPEG.PRS

Text File | 1993-12-20 | 22KB | 353 lines

                                        ôçä àöïï Åêéôöæä Äì ëÅäå ╬╠╬╠╬╠╧╧╬╠╡
         σΣα≤⌠±Σ α±≤ΦΓδΣ
                                   │           ûçÇô êÆ ëÅäå?
       ┌┐                          │
       └┘                          │     JPEG (pronounced "jay-peg") is
       ┌┐┌────┐┌────┐┌────┐        │  a standardized  image compression
       │││ ┌┐ ││ ── ││ ┌┐ │        │  mechanism.
       │││ └┘ ││ ───┤│ └┘ │        │     JPEG    stands    for    Joint
       │││ ┌──┘└────┘├──  │        │  Photographic  Experts  Group, the
       ││└─┘         └────┘        │  original  name  of  the committee
       ││  ß√ ≤ε∞ δαφΣ             │  that wrote the standard.
      ┌┘│   Organizer              │
      │ │  Independent             │     JPEG    is     designed    for
      └─┘  JPEG  Group             │  compressing either  full-color or
                                   │  gray-scale  images   of  natural,
                                   │  real-world scenes.  It works well
  " ÷τΣφ α Θ∩Σµ σΦδΣ Φ≥ ∞απΣ σ±ε∞  │  on    photographs,   naturalistic
σ⌠δδ-Γεδε±  πα≤α,  ≤τΣ  Θ∩Σµ ÷Φδδ  │  artwork,  and  similar  material;
≤√∩ΦΓαδδ√ ßΣ α  σαΓ≤ε± εσ σε⌠± ε±  │  not so well  on lettering, simple
σΦ⌡Σ ≥∞αδδΣ± ≤ταφ α µΦσ σΦδΣ ∞απΣ  │  cartoons, or line drawings.  JPEG
σ±ε∞ ≤τΣ ≥α∞Σ πα≤α. "              │  handles  only  still  images, but
                                   │  there  is   a   related  standard


called MPEG for motion pictures.   │  JPEG may  be  a problem  for you,
                                   │  even if they are invisible to the
   JPEG is  "lossy," meaning that  │  eye.
the   decompressed   image  isn't  │
quite the  same  as  the  one you  │     A useful  property of  JPEG is
started   with.      (There   are  │  that the degree  of lossiness can
lossless     image    compression  │  be     varied     by    adjusting
algorithms,  but   JPEG  achieves  │  compression  parameters.     This
much greater  compression than is  │  means that  the  image  maker can
possible with  lossless methods.)  │  trade  off   file   size  against
JPEG is designed to exploit known  │  output image  quality.    You can
limitations  of  the  human  eye,  │  make *extremely*  small  files if
notably the fact that small color  │  you don't mind poor quality; this
details aren't  perceived as well  │  is useful  for  applications like
as  small  details  of light-and-  │  indexing      image     archives.
dark.  Thus, JPEG is intended for  │  Conversely, if  you  aren't happy
compressing images  that  will be  │  with the  output  quality  at the
looked at by humans.  If you plan  │  default compression  setting, you
to  machine-analyze  your images,  │  can jack up the quality until you
the  small  errors  introduced by  │  are satisfied,  and accept lesser


compression.                       │  to one.)
                                   │
          ûçÿ öÆä ëÅäå?            │     If   your   viewing   software
                                   │  doesn't  support  JPEG  directly,
   There are two good reasons: to  │  you'll have  to  convert  JPEG to
make  your  image  files smaller,  │  some other format  for viewing or
and  to   store  24-bit-per-pixel  │  manipulating images.  Even with a
color data  instead of 8-bit-per-  │  JPEG-capable  viewer,   it  takes
pixel data.                        │  longer to decode  and view a JPEG
                                   │  image than to view  an image of a
   ∞αΩΦφµ Φ∞αµΣ  σΦδΣ≥ ≥∞αδδΣ± is  │  simpler  format   such   as  GIF.
a big win  for transmitting files  │  Thus, using JPEG is essentially a
across networks and for archiving  │  time/space tradeoff:  you give up
libraries of images.   Being able  │  some time  in  order to  store or
to compress a  2 Mbyte full-color  │  transmit an image more cheaply.
file down  to  100  Kbytes  or so  │
makes a  big  difference  in disk  │     It's  worth  noting  that when
space and transmission time!  (If  │  network or  phone transmission is
you are  comparing GIF  and JPEG,  │  involved, the  time  savings from
the size ratio  is more like four  │  transferring a  shorter  file can


be greater  than  the  extra time  │  seem  as  obsolete  as black-and-
needed to decompress the file.     │  white MacPaint format does today.
                                   │  Furthermore, for reasons detailed
   ≤τΣ     ≥ΣΓεφπ     σ⌠φπα∞Σφ≤αδ  │  later, JPEG  is  far  more useful
απ⌡αφ≤αµΣ  εσ  Θ∩Σµ  Φ≥  ≤τα≤  Φ≤  │  than  GIF  for  exchanging images
≥≤ε±Σ≥ σ⌠δδ Γεδε± Φφσε±∞α≤Φεφ: 24  │  among people  with widely varying
bits/pixel  (16  million colors).  │  display hardware.   Hence JPEG is
GIF,  the   other   image  format  │  considerably   more   appropriate
widely used  on Usenet,  can only  │  than  GIF  for  use  as  a Usenet
store 8 bits/pixel  (256 or fewer  │  posting standard.
colors). GIF  is  reasonably well  │
matched  to  inexpensive computer  │     A lot of people are scared off
displays --- most run-of-the-mill  │  by the  term "lossy compression".
PCs can't  display more  than 256  │  But when it comes to representing
distinct  colors   at  once.  But  │  real-world  scenes,  *no* digital
full-color  hardware  is  getting  │  image format  can retain  all the
cheaper all  the  time,  and JPEG  │  information that impinges on your
images  look  *much*  better than  │  eyeball.  In  comparison with the
GIFs on such  hardware.  Within a  │  real-world scene,  JPEG loses far
couple of  years, 8-bit  GIF will  │  less information  than GIF.   The


technical meaning  of "lossy" has  │  without  visible  loss,  bringing
nothing to do  with this, though;  │  the effective storage requirement
it refers to  loss of information  │  down to 1 to  2 bits/pixel.  30:1
over repeated compression cycles,  │  to 50:1  compression  is possible
a problem that you may or may not  │  with small  to  moderate defects,
care about.                        │  while     for    very-low-quality
                                   │  purposes  such   as  previews  or
       çÄû  ûäïï  âÄäÆ  ëÅäå       │  archive       indexes,      100:1
       éÄîÅæäÆÆ  êîÇåäÆ?           │  compression  is  quite  feasible.
                                   │  An  image  compressed  100:1 with
   Very well indeed, when working  │  JPEG takes up the same space as a
with its  intended type  of image  │  full-color        one-tenth-scale
(photographs and  suchlike).  For  │  thumbnail image,  but  it retains
full-color       images,      the  │  much  more  detail  than  such  a
uncompressed data  is normally 24  │  thumbnail.
bits/pixel.     The   best  known  │
lossless compression  methods can  │     For comparison,  a GIF version
compress such  data about  2:1 on  │  of the same image would start out
average.     JPEG  can  typically  │  by sacrificing most  of the color
achieve 10:1  to 20:1 compression  │  information to  reduce  the image


to  256  colors  (8  bits/pixel).  │  Because  the  human  eye  is much
This  provides  3:1  compression.  │  more   sensitive   to  brightness
GIF    has    additional    "LZW"  │  variations     than     to    hue
compression  built  in,  but  LZW  │  variations, JPEG can compress hue
doesn't work very well on typical  │  data more heavily than brightness
photographic  data;  at  most you  │  (gray-scale) data.
may get  5:1 compression overall,  │     A  gray-scale   JPEG  file  is
and it's not  at all uncommon for  │  generally   only   about  10%-25%
LZW to be  a net  loss (less than  │  smaller  than  a  full-color JPEG
3:1 overall compression).          │  file of  similar  visual quality.
   When a JPEG  file is made from  │  But  the  uncompressed gray-scale
full-color data,  using a quality  │  data  is  only  8  bits/pixel, or
setting  just   high   enough  to  │  one-third the  size of  the color
prevent  visible  loss,  the JPEG  │  data,    so     the    calculated
will  typically  be  a  factor of  │  compression ratio  is much lower.
four or  five smaller  than a GIF  │  The threshold of  visible loss is
file made from the same data.      │  often around  5:1 compression for
                                   │  gray-scale images.
   Gray-scale   images   do   not  │     The exact  threshold  at which
compress by  such  large factors.  │  errors become  visible depends on


your  viewing  conditions.    The  │  chrominance subsampling, discrete
smaller an  individual pixel, the  │  cosine   transforms,  coefficient
harder it is to  see an error; so  │  quantization,   and   Huffman  or
errors  are  more  visible  on  a  │  arithmetic entropy  coding.  This
computer  screen   (at  maybe  70  │  article's  too  long  already, so
dots/inch) than on a high-quality  │  I'm not  going  to say  more than
color  printout   (300   or  more  │  that here.
dots/inch).   Thus    a   higher-  │
resolution  image   can  tolerate  │  ûçÇô'Æ  Çïï   ôçêÆ  çÄÄÅïÇ  ÇüÄöô
more  compression  ...  which  is  │  éÄïÄöæ ÉöÇìôêÖÇôêÄì?
fortunate  considering  it's much  │
bigger to start with. The numbers  │     Most people  don't  have full-
quoted  above   are  typical  for  │  color (24 bit  per pixel) display
screen viewing.   Also  note that  │  hardware.     Typical     display
the  threshold  of  visible error  │  hardware stores  8 or  fewer bits
varies somewhat across images.     │  per pixel, so  it can display 256
                                   │  or  fewer  distinct  colors  at a
      çÄû âÄäÆ ëÅäå ûÄæè?          │  time.   To  display  a full-color
                                   │  image, the  computer  must choose
   The  buzz-words  to  know  are  │  an     appropriate     set     of


representative colors and map the  │  algorithm is best for all images.
image into  these  colors.   This  │
process    is    called    "color  │     Since  JPEG  is  a  full-color
quantization". (This is something  │  format, converting  a  color JPEG
of a  misnomer; "color selection"  │  image  for  display  on 8-bit-or-
or "color  reduction" would  be a  │  less   hardware   requires  color
better term.    We're  stuck with  │  quantization.     The  speed  and
the standard usage though.)        │  image quality  of  a  JPEG viewer
                                   │  running  on   such  hardware  are
   Clearly, color quantization is  │  largely    determined    by   its
a lossy  process.   It  turns out  │  quantization  algorithm.   You'll
that for most images, the details  │  see  great   variation  in  image
of    the    color   quantization  │  quality  among  viewers  on 8-bit
algorithm have *much* more impact  │  displays, much  more  than occurs
on the  final image  quality than  │  on 24-bit displays.
do any errors  introduced by JPEG  │     On the other hand, a GIF image
itself (except at the very lowest  │  has already been quantized to 256
JPEG quality settings).  Making a  │  or fewer  colors.   (A GIF *does*
good color quantization algorithm  │  have a specific  number of colors
is a  black  art,  and  no single  │  in its  palette,  and  the format


doesn't  allow   more   than  256  │  Furthermore, if  the maker didn't
palette entries.)    GIF  has the  │  use    a    high-quality    color
advantage  that  the  image maker  │  quantization   algorithm,  you're
precomputes       the       color  │  out  of  luck  ---  the  image is
quantization,  so  viewers  don't  │  ruined.
have  to;  this  is  one  of  the  │
things  that   make  GIF  viewers  │     For this reason, JPEG promises
faster than  JPEG  viewers.   But  │  significantly     better    image
this is  also  the *disadvantage*  │  quality than  GIF  for  all users
of  GIF:  you're  stuck  with the  │  whose  machines  don't  match the
maker's  quantization.    If  the  │  image  maker's  display hardware.
maker  quantized  to  a different  │  JPEG's full  color  image  can be
number of  colors  than  what you  │  quantized to  precisely match the
can display,  you'll either waste  │  viewer's     display    hardware.
display  capability  or  have  to  │  Furthermore, you will  be able to
quantize again  to further reduce  │  take    advantage    of    future
the  number   of   colors  (which  │  improvements    in   quantization
results  in   much  poorer  image  │  algorithms  (there  is  a  lot of
quality than if you had quantized  │  active research in this area), or
once  from  a  full-color image).  │  purchase better display hardware,


to  get  a  better  view  of JPEG  │  ±Σ∩±Σ≥Σφ≤Φφµ Φ∞αµΣ≥ Φφ α ∞αΓτΦφΣ-
images you already  have.  With a  │  ΦφπΣ∩ΣφπΣφ≤ σα≥τΦεφ.
GIF,  you're   stuck  forevermore  │
with what was sent.                │     It's sometimes  thought that a
                                   │  JPEG   converted   from   a   GIF
   A  growing  number  of  people  │  shouldn't      require      color
have   better-than-8-bit  display  │  quantization.    This  is  false:
hardware  already:   15-bit  "hi-  │  even when you feed a 256-or-less-
color" PC  displays,  true 24-bit  │  color GIF  into JPEG,  what comes
displays   on   workstations  and  │  out of  the  decompressor  is not
Macintoshes,  etc.     For  these  │  256  colors,   but  thousands  of
people, GIF  is already obsolete,  │  colors.    This  happens  because
as it  cannot represent  an image  │  JPEG's  lossiness   affects  each
to the full capabilities of their  │  pixel  a  little  differently, so
display.   JPEG images  can drive  │  two  pixels   that  started  with
these    displays    much    more  │  identical  colors   will  usually
effectively.                       │  come out  with slightly different
                                   │  colors.  Each original color gets
   Φφ  ≥τε±≤,  Θ∩Σµ  Φ≥  αφ  αδδ-  │  "smeared" into a  group of nearby
α±ε⌠φπ ßΣ≤≤Σ± ΓτεΦΓΣ ≤ταφ µΦσ σε±  │  colors.    Therefore quantization


is always  required to  display a  │  or  gray-scale,   but  number  of
color  JPEG   on   a  colormapped  │  colors   just   isn't   a  useful
display, regardless  of the image  │  concept for  JPEG, any  more than
source.                            │  it is for a real photograph.
                                   │
   The  same   effect   makes  it  │    ûçäì ÆçÄöïâ ê  öÆä  ëÅäå, Çìâ
nearly meaningless  to talk about  │    ûçäì ÆçÄöïâ ê Æôêéè ûêôç åêà?
the number  of  colors used  by a  │
JPEG image.  Even  if you were to  │     JPEG   is   *not*   going   to
count  the   number  of  distinct  │  displace GIF  entirely;  for some
pixel   values,   different  JPEG  │  types of images,  GIF is superior
decoders would give you different  │  in image  quality, file  size, or
results because of roundoff error  │  both.  One of the first things to
differences.   I occasionally see  │  learn about  JPEG is  which kinds
posted images  described as "256-  │  of images to apply it to.
color JPEG".   This tells me that  │
the poster  (a) hasn't  read this  │     Generally  speaking,  JPEG  is
FAQ  and  (b)  probably converted  │  superior to GIF for storing full-
the JPEG from a GIF.               │  color  or  gray-scale  images  of
JPEGs can be  classified as color  │  "realistic"  scenes;  that  means


scanned  photographs  and similar  │  visible defects. (One implication
material.        Any   continuous  │  of  this  is  that  large single-
variation  in   color,   such  as  │  color borders are  quite cheap in
occurs in  highlighted  or shaded  │  GIF files,  while  they  are best
areas, will  be  represented more  │  avoided in JPEG files.)
faithfully and  in less  space by  │
JPEG than by GIF.                  │     Computer-drawn   images  (ray-
   GIF does  significantly better  │  traced   scenes,   for  instance)
on  images   with   only   a  few  │  usually fall  between photographs
distinct  colors,  such  as  line  │  and   cartoons    in   terms   of
drawings  and   simple  cartoons.  │  complexity.  The more complex and
Not only is GIF lossless for such  │  subtly  rendered  the  image, the
images, but  it  often compresses  │  more  likely  that  JPEG  will do
them more  than  JPEG  can.   For  │  well on  it.  The  same  goes for
example,  large  areas  of pixels  │  semi-realistic  artwork  (fantasy
that are  all *exactly*  the same  │  drawings and such).
color    are    compressed   very  │
efficiently indeed by  GIF.  JPEG  │     Θ∩Σµ τα≥ α τα±π ≤Φ∞Σ ÷Φ≤τ ⌡Σ±√
can't squeeze  such data  as much  │  ≥τα±∩ ΣπµΣ≥: a  row of pure-black
as GIF  does  without introducing  │  pixels adjacent to a row of pure-


white pixels, for example.  Sharp  │  16  gray  levels  before  JPEG is
edges tend  to  come  out blurred  │  useful for gray-scale images.  It
unless  you   use  a   very  high  │  should also be  noted that GIF is
quality  setting.     Edges  this  │  lossless for gray-scale images of
sharp   are   rare   in   scanned  │  up to  256 levels,  while JPEG is
photographs,   but   are   fairly  │  not.
common  in  GIF  files:  borders,  │
overlaid   text,    etc.      The  │     If you have a large library of
blurriness     is    particularly  │  GIF images, you  may want to save
objectionable  with  text  that's  │  space by  converting the  GIFs to
only a few  pixels high.   If you  │  JPEG.   This is  trickier than it
have a GIF  with a  lot of small-  │  may seem  --- even  when the GIFs
size  overlaid  text,  don't JPEG  │  contain photographic images, they
it.                                │  are  actually  very  poor  source
                                   │  material  for  JPEG,  because the
   Plain   black-and-white   (two  │  images  have  been color-reduced.
level)  images  should  never  be  │  Non-photographic   images  should
converted to  JPEG;  they violate  │  generally be left in GIF form.
all  of   the   conditions  given  │     Good-quality photographic GIFs
above.   You need  at least about  │  can often  be  converted  with no


visible quality loss, but only if  │  have claimed.   The quality scale
you know  what you  are doing and  │  is purely  arbitrary; it's  not a
you take the time to work on each  │  percentage of anything.
image  individually.    Otherwise  │
you're likely  to  lose a  lot of  │     In fact, quality scales aren't
image quality  or waste  a lot of  │  even  standardized   across  JPEG
disk  space  ...  quite  possibly  │  programs.   The  quality settings
both.  See  later sections.        │  discussed in  this  article apply
                                   │  to   the   free   JPEG   software
     ûçÇô Çæä  åÄÄâ "ÉöÇïêôÿ"      │  described in  the  next  edition,
        ÆäôôêìåÆ àÄæ ëÅäå?         │  , and to  many  programs based on
                                   │  it. Other JPEG   implementations,
   Most JPEG  compressors let you  │  notably  Apple's  and HSI's,  use
pick  a   file  size   vs.  image  │  completely    different   quality
quality tradeoff  by  selecting a  │  scales;   for  instance,  Apple's
quality setting.   There seems to  │  scale covers 0-4, not 0-100. Some
be widespread confusion about the  │  programs  don't  even  provide  a
meaning   of    these   settings.  │  numeric scale,just "high"/"medium
"Quality 95" does  NOT mean "keep  │  /"low"-style choices.
95% of the  information", as some  │  (Fortunately,     this    doesn't


prevent different implementations  │  Q 75  first; if  you see defects,
from exchanging compressed files)  │  then go up.
                                   │     If  the  image  was  less than
   In most cases  the user's goal  │  perfect  quality  to  begin with,
is  to  pick  the  lowest quality  │  you might be able to drop down to
setting, or  smallest  file size,  │  Q   50    without   objectionable
that decompresses  into  an image  │  degradation.  On  the other hand,
indistinguishable     from    the  │  you  might   need  to   go  to  a
original.  This setting will vary  │  *higher* quality setting to avoid
from  one  image  to  another and  │  further loss. Q 85 to 95 is often
from one observer to another, but  │  best  for  converting  GIFs.
here are some rules of thumb.      │
                                   │     Except     for    experimental
   For  good-quality,  full-color  │  purposes, never go  above about Q
source   images,    the   default  │  95; using  Q  100 will  produce a
quality setting  (Q  75)  is very  │  file two or  three times as large
often  the   best   choice.  This  │  as Q 95, but of hardly any better
setting is  about the  lowest you  │  quality.  If you  see a file made
can go  without expecting  to see  │  with Q  100,  it's a  pretty sure
defects in a  typical image.  Try  │  sign that  the maker  didn't know


what he/she was doing.             │  CONVERTING  GIF  IMAGE S TO JPEG?
                                   │  DOES    LOSS    ACCUMULATE   WITH
   If you want  a very small file  │  REPEATED        COMPRESSION   AND
(say  for   preview  or  indexing  │  DECOMPRESSION?
purposes)  and  are  prepared  to  │  WHY ALL  THE ARGUMENT  ABOUT FILE
tolerate  large   defects,   a  Q  │  FORMATS?
setting in the  range of  5 to 10  │  HOW  DO  I  RECOGNIZE  WHICH FILE
is about right.  Q 2 or so may be  │  FORMAT I HAVE,  AND WHAT  DO I DO
amusing as "op art".               │  ABOUT IT?
                                   │  ISN'T THERE A LOSSLESS JPEG?
                                   │  WHAT ABOUT ARITHMETIC CODING? ñ
In the next edition, we will look  │
at:                                │
                                   │
WHERE  AND  HOW  CAN  I  GET JPEG  │
SOFTWARE?                          │
VIEWERS,   APPLICATION  PROGRAMS,  │
ETC.,   SOURCE   CODE  αφπ   σ±ΣΣ  │
Θ∩Σµ Γε∞∩±Σ≥≥ε±/πΣΓε∞∩±Σ≥≥ε±.      │
WHAT ARE SOME  RULES OF THUMB FOR  │