Abstract
A modular library for digital halftoning methods. The aim were photorealistic methods utilizing space-filling curves and blue-noise, artistic methods were considered as well. From the point of view of software design, emphasis was on modularity and extensibility, in order to support experimenting, even with new algorithms. The library along with a graphical user interface has been integrated as a plug-in with the framework of a popular image-processing application GIMP.
- Author: Bohumir Zamecnik
- Release Date: 2009-08-03
- License: The MIT License, see the LICENSE file
- Project website (on SourceForge):
- Thesis text: Advanced Halftoning Methods
Halftone Laboratory should run everywhere where Gimp# is able to run. Currently, it was tested sucessfully on Windows XP SP2 with .NET 3.5, GIMP 2.4.6, GIMP# 0.14 and GTK# 2.10.
- Operating system: Windows XP, Linux
- Microsoft .NET Framework 3.5 / Mono 2.4
- GIMP >= 2.4
- GIMP# >= 0.13
First make sure you have Microsoft .NET Framework 3.5 installed, otherwise HalftoneLab will not work. You can eventually use the enclosed incremental installers of .NET 2.0, 3.0, 3.5, and optionally 3.5 SP1. Note that some of those installer might need internet accesss.
Then GIMP and GIMP# must be installed, run gimp-2.4.6-i686-setup.exe and
then gimp-sharp-setup-0.14-gimp-2.4.exe. After that you can install
HalftoneLab with halftonelab-setup.exe. All installers act as wizards, so the
installation is straightforward.
HalftoneLab stores its configuration in halftonelab.cfg file, located in the Application Data directory. On Windows platform it can be Application Data in the user directory, on other systems .config in the home directory, but it depends on specific operating system and its configuration. If the configuration file does not exist, it is automatically created and filled with some example configuration.
The basic graphical user interface for HalftoneLab project is available in form of a GIMP plug-in. So at first launch the GIMP application and open an image. Currently, HalftoneLab is able only to process GIMP images in greyscale mode. To convert an image to that mode click in the menu on Image -> Mode -> Greyscale. Then you can start HalftoneLab plug-in: Filters -> Distorts -> Halftone Laboratory.
A main dialog opens. It consist of several panels:
- Configuration saving and loading
- Pre-processing
- Halftone method
- Post-processing
This panel enables you to manage algorithm configurations - save current one
along with a name and description to a persistent storage and load it again.
Select a name from the combobox to load that configuration. There are two
special configurations: _DEFAULT reverts all modules to their default
settings, _LAST is the last configuration automatically saved at the end of
the previous session.
To save the current configuration hit the Save button on the configuration
panel. A dialog prompting for configuration name and description appears. A
name is mandatory and cannot be "_DEFAULT" or "_LAST" or the configuration is
not saved, a description is optional, but can help to quickly see what the
algorithm does. The configurtion selected in the combobox can be removed
hitting the Delete button.
In the pre-preprocessing panel you can configure how the image will be altered before halftoning itself takes place. Modules are ordered as they will be exeuted. Each module can be enabled or disabled.
With this module you can scale the image by a given factor using selected interpolation method - Nearest neighbour, Bilinear, Bicubic, Lanczos.
Sharpening prior to halftoning can help preserving image details as most halftoning methods slightly blur the image. The amount of sharpening can be set.
Dot gain (see page dotgain) caused by the behaviour of some printing processes can be corrected here. Simple gamma correction acts as a rough approximation of dot gain curves.
Post-processing panel embraces modules executed after the halftoning. It acts similarly to the pre-processing panel.
The post-processing resize module acts the same way as its pre-processing counterpart. It is typically used to down-sample the image. You can perform supersampling technique with the two resize modules - in the pre-processing one set the desired upsample factor and in the other module enable Supersampling checkbutton. The downsample factor will be computed automatically. Bilinear interpolation is a recommended method for supersampling.
If you want to smoothen the halftoned image without the computation overhead of supersamping you can use the Smoothen module. First it Gaussian-blurs the image with given radius and then it applies Levels GIMP command. Thus, the rough black and white contours get perfectly smooth. However, the cost is bluring some fine details.
This is the most important control panel in the plug-in. There you can set the type of halftoning method to be used and its details. The panel itself follows a concept of submodule selector, which appears on many places inside specific module configuration dialogs. It consists of three elements: In a combobox there are submodule types, one of which can be selected. On hitting the Edit button a configuration dialog for that submodule type opens (provided there is anything to configure). Sometimes, the submodule is optional and it is possible that no type is set. To achieve this there is a null checkbutton. If it gets checked the module is unset and its internal settings get deleted.
In thresholding methods the image is processed pixel-wise along a scanning
order given by a ScanningOrder module. Each pixel is quantized using threshold
computed by a ThresholdFilter module and the quantization error is optionally
diffused by a ErrorFilter module. Each submodule can be configured via a
submodule configuration panel. As for ErrorFilter, there is a Use error
filter? checkbutton to enable or disable the module without removing its
configuration.
SFC clustering method, described on page sfc:clustering, works differently. The main parameter, Maximum cell size, controls the coarseness of the halftone (or the spatial vs. tonal resolution trade-off) - lower values give finer detais, but higher values enable representing more tones. There are checkbuttons for controlling the use of adaptive clustering (ie. varying cell sizes according to local detail amount) and cluster positioning techniques. Setting Minimum cell size value is meaningful only when using adaptive clustering.
The type of space-filling curve to scan along can be selected via Scanning
order submodule panel. Currently only Hilbert SFC is supported. An optional
VectorErrorFilter can be used.
Matrix threshold filter enables you to edit a single thresholding matrix. The matrix dimensions can be changed using the Resize button.
Dynamic matrix threshold filter allows a different matrix for each pixel intensity or a range of intensities as well as perturbing their threshold coefficients with random noise. There is a table of records consisting of a threshold matrix, starting intensity of that rangethe range spans up to the next record and noise amplitude. You can add a new record hitting the New button, which opens a separate dialog. Selected record can be edited using the Edit button in the same dialog or deleted with the Delete button. To delete all records at once hit the Clear all button. Applying perturbation noise can be enabled or disabled for the whole table with the Noise enabled? checkbutton.
A spot function analytically defines threshold values for digital screening (see page spotfunc). There are several presets for most common spot functions: Euclid dotwith growth: circle -> square at intensity 0.5 -> circle, Euclid dot with random perturbation, Square dot, Line, Triangle dot. For each preset two parameters can be set: Screen angle (in radians) - the angle of screen rotation, Screen line distance (in pixels) - distance between adjacent screen elements.
Image threshold filter behaves much like {Spot function threshold filter, except that spot functions are not evaluated at run-time but rather a big threshold matrix is pre-generated - to a GIMP image. The advantage is that the image can be then distorted with GIMP image filters. The possibilities are tremendous. A few examples are provided in the form of presets. The initial spot function can be configured the same way as in Spot function threshold filter. Currently, the effects applied cannot be controled, only a description is shown. One of the presets use GIMP Patterns to tile the plane, histogram is then equalized - this with an error filter enabled give approximately the Veryovka-Buchanan method Veryovka99 veryovka. To select a pattern, go to Dialogs -> Pattern in Image window menu and select the desired pattern.
Matrix error filter performs error-diffusion with a single error matrix. The way of its editing is similar to Matrix threshold filter dialog. Dimensions of the matrix can be changed using the Resize button.
Error matrix coefficients are represented as fractions - numerators are in the
table and a common denominator is in the Divisor spinbutton. As the sum of all
error matrix coefficients must be equal to
The relative position of source pixel in the matrix can be controlled by the Source offset X (the Y offset is always the same - the first row).
This is a concept similar to Dynamic matrix threshold filter (and the controls are almost identical) - there can be different error matrices for different pixel intensities or intensity ranges, however, there is no noise added.
Error matrix coefficients can be generated randomly for each pixel. There is a 'template' matrix which defines the dimensions of the final matrix and where the coefficients would be generated. A constraint is that error can be distributed only to places after the source pixel (following the scanning order). There are two possible modes controlled by the Randomize coefficient count? checkbox: If it is enabled, coefficients will be generated exactly in place of all non-zero coefficients in the template matrix, maintaining a constant number of them, otherwise the number and position of coefficients will be varied up to the available capacity of the matrix.
For perturbing error matrix coefficients with random noise a separate filter is used. Inside it has a MatrixErrorFilter child filter as a submodule controlled by a standard submodule panel. In addition to it the amount of noise can be controlled by the Perturbation amplidute scale.
Currently, Perturbed matrix error filter does not support Dynamic matrix error filter as its child filter.
For 1-D scanning, such as along an space-filling curve, a vector error filter is suitable. Its interface is similar to that of Matrix error filter, except that it can be resized only in one dimension (Length).
For general scanning order you can select among Scanline, Serpentine and Hilbert scanning, for SFC scanning only Hilbert is available. Scanline processes pixels line by line in the same direction, serpentine in a zig-zag direction. Hilbert scans along an approximation of the Hilbert space-filling curve.