Der Mediaplex Sampler - Die 6 von Plex/6_v_plex.zip/6_v_plex/DISK2/MULTI_04/CONVRT.ZIP/CONVERT.DOC

Text File | 1993-05-12 | 12KB | 244 lines

                Convert version 1.2
                   May 12, 1993

    Copyright 1993 Stephen B. Coy
    All Rights Reserved

        Unmodified copies of Convert may be freely distributed via
    BBSs, Internet and services such as CompuServe and BIX.  You may
    not include Convert as part of any commercial product, that
    includes shareware, without explicit permission.  I can be
    contacted at the address below for more info and bug reports.

        Stephen Coy
        Vivid Software
        15205 NE 13th Pl., #2904
        Bellevue, WA  98007

        [phone]

        CompuServe : 70413,136
        Internet   : [e-mail]

        Convert is based on Convert.exe which is included with
    the shareware ray tracer Vivid.  (plug, plug!)  I've decided to
    move away from Vivid's own .img format and join the rest of the
    world using Truevision's Targa format.  Convert reads truecolor
    Targa files (16, 24 and 32 bit) and Vivid's .img files and
    converts them to 8-bit GIFs.  When 32-bit Targa files are
    converted the alpha data is ignored.  Convert will read both
    run-length encoded and unencoded Targas.

         Convert.exe is a real-mode version which should run on
    just about any system.  Conv_386.exe is a 386 compiled version
    which runs about twice as fast and uses a DOS extender so it
    shouldn't run out of memory.

         When running convert you have two main decisions to make: a)
        how should the program pick the palette and b) should the program
        do any dithering of the output.  Picking the palette is not
        always an easy problem.  Standard VGA allows us to view only 256
        of the available 262,144 colors.  An "average" lores image will
        have about 7000 colors which we need to represent with our
        palette of 256.  A higher resolution image my have tens of
        thousands of colors to represent.  The following algorithms are
    currently implemented in convert.

             Palette File -- A palette may be input via a .map file.
    This uses the same format as .map files created/read by Fractint.
        This option is useful for creating multiple gifs with the same
        palette for animation.

             Popularity -- The popularity algorithm chooses the palette
        by picking the 256 colors which represent then greatest number of
        pixels in the image.  The remaining colors in the image are
        mapped onto the palette using a minimum distance formula.  The
        popularity algorithm tends to work best on images with a lower
        number of colors.  One typical problem of the algorithm is that
        small highlights may not be colored correctly.  Highlights
        generally only cover a few pixels so their color usually doesn't
        have enough representation in the image to be chosen by the
    popularity algorithm.  To help alleviate this problem, Convert
        forces the corners of the color cube (white, red, green, blue,
        cyan, magenta, yellow, and black) to be selected as the first
        eight palette entries.  Since most highlights are white this
        greatly helps reduce unwanted artifacts in the resulting image.
    This is the default option for choosing the palette.

             Median Cut -- The median cut algorithm tries to choose a
        more well balanced set of colors to represent the image.  The
        general idea of the median cut is to choose a palette in which
        each entry represents about the same number of pixels in the
        image.  This helps to correctly color highlights that a pure
    popularity algorithm might miss.  Convert also allows you to
        limit the number of pixels represented by any one color.  This,
        in effect, increases the importance of the colors in highlights
        and other small regions of the image.  With this limit set to
        one, every color in the image is given the same weight without
    regard for the number of pixels it covers.  This is the option
    you will want to use 90% of the time.

             Fixed Palette -- The fixed palette option uses a
        predetermined palette instead of choosing one based on the
        content of the image.  This has the advantage of being much
    faster.  When choosing by popularity or median cut Convert must
        first build a tree structure in memory containing every color in
        the image and a count of the number of pixels represented by that
        color.  After choosing the palette the colors in the tree must
        then be mapped onto the colors in the palette.  This can slow
    Convert down quite a bit.  When using a fixed palette the image
        colors are mapped directly onto the palette colors via simple
        equations eliminating the need for the tree structure and the
        costly remapping of the images colors to an arbitrary palette.
    Also, the current version of Convert can only support a tree
        containing about 57,000 colors.  After this my clone runs out of
    RAM.  The fixed palettes supported by Convert are:

             0 : A grey scale image using 256 grey scale tones.  Due to
        VGA limitations this will display as 64 shades on most PC
    systems.  Of the three fixed palettes supported by Convert this
    is the only one I ever use.

             1 : Divides the color cube into 8 shades of red, 8 shades of
        green, and 4 shades of blue.  Generally produces fairly bad
        images but kept in for nostalgia.  Blue was chosen to get the
        lower resolution because in general the human eye is much less
        sensitive in the blue range and has a harder time focusing on
        edges therefore the loss of color resolution is not missed as
        much.

             2 : Divides the color cube into 6 shades of red, 7 shades of
        green, and 6 shades of blue.  This option gives the best balance
        of speed vs good color representation.  I've found that it works
        best with images that have a large number of colors.  The
        addition of dithering is usually helpful.

             Dithering is the process of displaying pixels of differing
        colors next to each other in such a way as to produce the
        illusion of more colors.  An every day example of this is your
        television.  Stick your nose up to the picture tube and notice
        that the screen is made up of tiny red, green, and blue dots
        which vary in intensity.  These are the only colors your TV
        actually produces but your eye combines them to produce the rest
        of the spectrum.  Well, at least the spectrum according to NTSC.

             Floyd Steinberg dithering is the original error diffusion
        dithering algorithm.  Error diffusion dithering attempts to
        compensate for a limited palette by insuring that the sum of the
        errors in any region of the image is zero even though the colors
        at each individual pixel may contain some error.  At each pixel
    as Convert scans through the image the nearest color for that
        pixel is chosen.  Generally this color will not be exact.  The
        difference between the chosen palette color and the true color is
        then added to the neighboring pixels which have not been scanned
        yet.  Floyd Steinberg dithering has the advantage of having a
        "random" look to it.  Generally, it does not produce noticeable
        patterns in the output image.

             Ordered dithering works by adding a fixed amount to each
        pixel based on its location.  The sum of these additions over a
        region of the image is equal to zero.  Since ordered dither does
        not take into account the values of neighboring pixels it
        produces fixed patterns in the images which may be distracting in
        some cases.  Ordered dithering combined with a fixed palette has
    an advantage if you are producing single fames for an animation.
    During an animation the background (non-moving) regions of the
    image will stay constant whereas if you use Floyd Steinberg
    dithering the patterns constantly shift around and below any
    moving objects on the screen.

             The random noise option allows you to add white noise to an
        image.  In some cases this may help as much as dithering.  It can
        also be used in conjunction with either of the dithering options.
        Interesting effects may be obtained by increasing the range of
        the noise.  Images converted with a large amount of noise tend to
        take on a distinctly grainy look.  Interesting, but not overly
    useful.  The default range for the noise is +-4 on a 0..255
        range.  This value was chosen strictly because it seemed to look
        good in most cases.  Random noise has the advantage of being
        constant from image to image.  This means that smooth areas in a
        series of images will not shimmer during animation.

         Ok, fine, now how do I run Convert? Well, it goes like this:

    C> convert [-m #] [-f #] [-p palfile] [-d] [-o] [-r [#]] file

    If the file name has no extension then Convert will first try
    and open file.tga and then file.img.  If the filename already
    has an extension then it is used "as-is."

    [-m #] chooses median cut and limits each color to represent
        at most # pixels.  # defaults to 64K and is optional.

    [-f #] chooses fixed palette #.  # defaults to 0, grey
        scale.

    [-p palfile] reads the palette out of the file palfile.map.
        If this file does not exist the palette will be chosen by
        median cut, popularity, or fixed depending on what other
        flags are set.  After the palette is chosen it will then be
        written to the file palfile.map.

    [-d] chooses Floyd Steinberg dithering.

    [-o] chooses ordered dithering.

    [-r #] chooses random noise +- #.  # defaults to 4.

    For example:

        C> convert -f 2 -d test

    will convert test.img to test.gif using fixed palette 2 and
        Floyd Steinberg dithering.

    Assuming that test.map does not exist

        C> convert -m -p test test1
        C> convert -m -p test test2
        C> convert -m -p test test3
        C> convert -m -p test test4

        will convert test1 using median cut to choose the palette.  The
        palette will then be saved to test.map.  Test2, test3, and test4
        will then be converted using the palette in test.map.

             In general the best results will be obtained by using the -m
    -d flags as in Convert -m -d test.  For images that tend to be
    fairly monochromatic I've found that -m -d -r works well.
        Adding the small amount of random noise helps break up some of
        the banding.  On the down side it also adds to the .gif file size
        since it is tough to compress random noise well.  To get nice
    looking grey scale images try -f 0 -r.

        Questions, problems and suggestions can be addressed to
    me at the address at the top of this file.  I can also be
    found on PCGnet, Internet and CompuServe.

                Have fun,

                    Stephen


    Trademarks:

        BIX .......... Byte
        GIF .......... CompuServe
        Targa ........ Truevision
        Vivid ........ Vivid Software

version 1.0
    Original version

version 1.1
    Fixed bug with RLE encoded Targa files having the red and blue
    planes swapped.

version 1.2
    Added support for reading 16-bit Targa files.

    Added 5-bit prequantizing using te -5 flag on the command
    line.  This is most effective in images with large numbers
    of colors.