NAME

Imager::Filters - Entire Image Filtering Operations

SYNOPSIS

  use Imager;

  $img = ...;

  $img->filter(type=>'autolevels');
  $img->filter(type=>'autolevels', lsat=>0.2);
  $img->filter(type=>'turbnoise')

  # and lots others

  load_plugin("dynfilt/dyntest.so")
    or die "unable to load plugin\n";

  $img->filter(type=>'lin_stretch', a=>35, b=>200);

  unload_plugin("dynfilt/dyntest.so")
    or die "unable to load plugin\n";

  $out = $img->difference(other=>$other_img);

DESCRIPTION

Filters are operations that have similar calling interface.

Types of Filters

Here is a list of the filters that are always avaliable in Imager. This list can be obtained by running the filterlist.perl script that comes with the module source.

  Filter          Arguments   Default value
  autolevels      lsat        0.1
                  usat        0.1
                  skew        0

  bumpmap         bump lightx lighty
                  elevation   0
                  st          2

  bumpmap_complex bump
                  channel     0
                  tx          0
                  ty          0
                  Lx          0.2
                  Ly          0.4
                  Lz          -1 
                  cd          1.0 
                  cs          40.0
                  n           1.3
                  Ia          (0 0 0)
                  Il          (255 255 255)
                  Is          (255 255 255)

  contrast        intensity

  conv            coef

  fountain        xa ya xb yb
                  ftype        linear
                  repeat       none
                  combine      none
                  super_sample none
                  ssample_param 4
                  segments(see below)

  gaussian        stddev

  gradgen         xo yo colors dist

  hardinvert

  mosaic          size         20

  noise           amount       3
                  subtype      0

  postlevels      levels       10

  radnoise        xo           100
                  yo           100
                  ascale       17.0
                  rscale       0.02

  turbnoise       xo           0.0
                  yo           0.0
                  scale        10.0

  unsharpmask     stddev       2.0
                  scale        1.0

  watermark       wmark
                  pixdiff      10
                  tx           0
                  ty           0

All parameters must have some value but if a parameter has a default value it may be omitted when calling the filter function.

A reference of the filters follows:

autolevels

scales the value of each channel so that the values in the image will cover the whole possible range for the channel. lsat and usat truncate the range by the specified fraction at the top and bottom of the range respectivly.

bumpmap

uses the channel elevation image bump as a bumpmap on your image, with the light at (lightx, lightty), with a shadow length of st.

bumpmap_complex

uses the channel channel image bump as a bumpmap on your image. If Lz<0 the three L parameters are considered to be the direction of the light. If Lz>0 the L parameters are considered to be the light position. Ia is the ambient colour, Il is the light colour, Is is the color of specular highlights. cd is the diffuse coefficient and cs is the specular coefficient. n is the shininess of the surface.

contrast

scales each channel by intensity. Values of intensity < 1.0 will reduce the contrast.

conv

performs 2 1-dimensional convolutions on the image using the values from coef. coef should be have an odd length.

fountain

renders a fountain fill, similar to the gradient tool in most paint software. The default fill is a linear fill from opaque black to opaque white. The points A(xa, ya) and B(xb, yb) control the way the fill is performed, depending on the ftype parameter:

linear: the fill ramps from A through to B.
bilinear: the fill ramps in both directions from A, where AB defines the length of the gradient.
radial: A is the center of a circle, and B is a point on it's circumference. The fill ramps from the center out to the circumference.
radial_square: A is the center of a square and B is the center of one of it's sides. This can be used to rotate the square. The fill ramps out to the edges of the square.
revolution: A is the centre of a circle and B is a point on it's circumference. B marks the 0 and 360 point on the circle, with the fill ramping clockwise.
conical: A is the center of a circle and B is a point on it's circumference. B marks the 0 and point on the circle, with the fill ramping in both directions to meet opposite.

The repeat option controls how the fill is repeated for some ftypes after it leaves the AB range:

none: no repeats, points outside of each range are treated as if they were on the extreme end of that range.
sawtooth: the fill simply repeats in the positive direction
triangle: the fill repeats in reverse and then forward and so on, in the positive direction
saw_both: the fill repeats in both the positive and negative directions (only meaningful for a linear fill).
tri_both: as for triangle, but in the negative direction too (only meaningful for a linear fill).

By default the fill simply overwrites the whole image (unless you have parts of the range 0 through 1 that aren't covered by a segment), if any segments of your fill have any transparency, you can set the combine option to 'normal' to have the fill combined with the existing pixels. See the description of combine in Imager/Fill.

If your fill has sharp edges, for example between steps if you use repeat set to 'triangle', you may see some aliased or ragged edges. You can enable super-sampling which will take extra samples within the pixel in an attempt anti-alias the fill.

The possible values for the super_sample option are:

none: no super-sampling is done
grid: a square grid of points are sampled. The number of points sampled is the square of ceil(0.5 + sqrt(ssample_param)).
random: a random set of points within the pixel are sampled. This looks pretty bad for low ssample_param values.
circle: the points on the radius of a circle within the pixel are sampled. This seems to produce the best results, but is fairly slow (for now).

You can control the level of sampling by setting the ssample_param option. This is roughly the number of points sampled, but depends on the type of sampling.

The segments option is an arrayref of segments. You really should use the Imager::Fountain class to build your fountain fill. Each segment is an array ref containing:

start: a floating point number between 0 and 1, the start of the range of fill parameters covered by this segment.
middle: a floating point number between start and end which can be used to push the color range towards one end of the segment.
end: a floating point number between 0 and 1, the end of the range of fill parameters covered by this segment. This should be greater than start.
c0
c1: The colors at each end of the segment. These can be either Imager::Color or Imager::Color::Float objects.
segment type: The type of segment, this controls the way the fill parameter varies over the segment. 0 for linear, 1 for curved (unimplemented), 2 for sine, 3 for sphere increasing, 4 for sphere decreasing.
color type: The way the color varies within the segment, 0 for simple RGB, 1 for hue increasing and 2 for hue decreasing.

Don't forget to use Imager::Fountain instead of building your own. Really. It even loads GIMP gradient files.

gaussian

performs a gaussian blur of the image, using stddev as the standard deviation of the curve used to combine pixels, larger values give bigger blurs. For a definition of Gaussian Blur, see:

  http://www.maths.abdn.ac.uk/~igc/tch/mx4002/notes/node99.html

gradgen

renders a gradient, with the given colors at the corresponding points (x,y) in xo and yo. You can specify the way distance is measured for color blendeing by setting dist to 0 for Euclidean, 1 for Euclidean squared, and 2 for Manhattan distance.

hardinvert

inverts the image, black to white, white to black. All channels are inverted, including the alpha channel if any.

mosaic

produces averaged tiles of the given size.

noise

adds noise of the given amount to the image. If subtype is zero, the noise is even to each channel, otherwise noise is added to each channel independently.

radnoise

renders radiant Perlin turbulent noise. The centre of the noise is at (xo, yo), ascale controls the angular scale of the noise , and rscale the radial scale, higher numbers give more detail.

postlevels

alters the image to have only levels distinct level in each channel.

turbnoise

renders Perlin turbulent noise. (xo, yo) controls the origin of the noise, and scale the scale of the noise, with lower numbers giving more detail.

unsharpmask

performs an unsharp mask on the image. This is the result of subtracting a gaussian blurred version of the image from the original. stddev controls the stddev parameter of the gaussian blur. Each output pixel is: in + scale * (in - blurred).

watermark

applies wmark as a watermark on the image with strength pixdiff, with an origin at (tx, ty)

A demonstration of most of the filters can be found at:

  http://www.develop-help.com/imager/filters.html

(This is a slow link.)

Plugins

It is possible to add filters to the module without recompiling the module itself. This is done by using DSOs (Dynamic shared object) avaliable on most systems. This way you can maintain our own filters and not have to get me to add it, or worse patch every new version of the Module. Modules can be loaded AND UNLOADED at runtime. This means that you can have a server/daemon thingy that can do something like:

  load_plugin("dynfilt/dyntest.so")
    or die "unable to load plugin\n";

  $img->filter(type=>'lin_stretch', a=>35, b=>200);

  unload_plugin("dynfilt/dyntest.so")
    or die "unable to load plugin\n";

Someone decides that the filter is not working as it should - dyntest.c modified and recompiled.

  load_plugin("dynfilt/dyntest.so")
    or die "unable to load plugin\n";

  $img->filter(%hsh);

An example plugin comes with the module - Please send feedback to addi@umich.edu if you test this.

Note: This seems to test ok on the following systems: Linux, Solaris, HPUX, OpenBSD, FreeBSD, TRU64/OSF1, AIX. If you test this on other systems please let me know.

Image Difference

You can create a new image that is the difference between 2 other images.

  my $diff = $img->difference(other=>$other_img);

For each pixel in $img that is different to the pixel in $other_img, the pixel from $other_img is given, otherwise the pixel is transparent black.

This can be used for debugging image differences ("Where are they different?"), and for optimizing animated GIFs.

Note that $img and $other_img must have the same number of channels. The width and heigh of $diff will be the minimum of each of the width and height of $img and $other_img.