Scenes Understanding and Color Spaces

S. Chastel and A. Trémeau

LIGIV - Université de Saint-Étienne
10, rue Barrouin
F-42000 SAINT-ÉTIENNE
e-mail: chastel,tremeau@vision.univ-st-etienne.fr

The Colorspace Tool

Colorspace is a tool that displays color information in different color spaces. It proposes two visualization methods: a 2D representation of the colors (image in false colors) and a 3D representation of them. We will mostly use that last kind of visualization.

This section will shortly introduce the basic commands that you will have to use. A complete documentation of colorspace¹ is avalaible at:

http://www.ligiv.org/people/colantoni/couleur.org/colorspace.html

(you may also open http://www.ligiv.org and follow the links people, colantoni, and Color Space).

For the IPCV'02 exercises, files have to be downloaded from:

http://www.uni-koblenz.de/~lb/lb_activities/ipcv02/download/LIGIV/index.html

You will have to unzip the file named colorspace.zip under a local directory where you have writing permissions (e.g. c:\temp). You should also download images from the same url to that local directory.

Opening an image

The selection of an image can be made by clicking ``Open'' in the ``File'' menu. Area selection (also named Regions Of Interest: ROI) can be made by clicking ``Selection'' and selecting the shape that you wish to use (lines, rectangles, hand...). Ensure that the ``Display selection'' toggle is activated. You may ``undo'' a selection thanks to the Ctrl-Z key sequence. The ``Color spaces'' entry on the main menu bar opens a window named ``Color space converter''.

The ``Color space'' selection is made in that window: sixteen different color spaces may be chosen. That window is the core of the application.

The Image entry

In that window, the ``Image'' entry allows the visualization of the different channels of the picture (``Visualization'', ``Channels''). By channels we mean the effective axis in the previously chosen color space. For instance, if the La*b* color space is chosen, channel 0 stands for L, channel 1 for a* and channel 2 for b* As the image will be displayed in false colors, specific LUT may also be chosen (``Visualization'', ``Channels'', ``LUT'').

The 3D Color Space entry

In the window ``Color space converter'', the entry ``3D Color Space'' allows the representation of the coordinates of colors of pixels in the corresponding color space. For spaces that use XYZ, the primaries and the white reference may be freely chosen (by default, it is CIE and a C illuminant). Visualizing the colors of the whole image is made by drag-and-dropping the image from the ``Display'' window to ''Color Space'' window with the mouse right button while holding the Ctrl and Shift keys pressed. The visualization of specific ROI can be made by drag-and-dropping from the ``Display'' window to ''Color Space'' window with the mouse right button while holding the Shift key pressed. ``Informations'' displays some statistical informations on the color cloud.

The 3D Histogram entry

In the window ``Color space converter'', the entry ``3D Histogram'' allows the representation of the histogram of the colors. The visualization process is the same that the one seen for ``3D Color Space''. By clicking ``Visualization'' and ``Color Properties'', sampling may be modified by chosing the number of different classes along the three axes.

Navigation

A 3D navigation can be made in the ''Color Space'' window and in the ``Histogram'' Window. That navigation process uses a camera which is defined by its position and the position of its target. Left button modifies the camera position and keeps the distance of the target (rotation around the target). Ctrl+Left button (or Middle button) modifies the distance to the target. Right button translate the camera position along an plane orthogonal to the line passing through the camera and the target.

Color and Objects Discrimination

Represent the colors of image image1 in different class of color spaces (technological, non-correlated, uniform...). You will use the ``Color spaces'' entry in the menu bar and in the ``Color space converter'' window the ``3D color space'' entry. You may also do that for selected particular areas in the image to display their own properties.

The aim here is to determinate the space that does the best discrimination between objects in the scene according to a ``segmentation'' point of view.

Color Space Projections

Use the four following images mandrill, lenna, image1, and synthese.

Using `3D color space'' (as in the previous exercise), try to find a plane in the colorspace such that the projections of color on that plane is discriminating.

In the same way, try to find an axis that should discriminate colors.

You may appreciate the rightness of your ideas with the ``Image'' entry in the ``Color space converter'' window.

The aim here is to determinate if there exist subsets in the picture that would be sufficient to discriminate between colors.

Color Histogram. Color Distance

Use the 4 following images mandrill, lenna, image1, and synthese.

Represent the histograms associated to these images. You will use the ``3D Histogram'' entry in the ``Color space converter'' window. The aspect is highly parametrable if you use the ``Color space configuration''window (``Visualization'', ``Color Properties'').

The aim here is to show the existence (or the inexistence) of sets of representative colors and to see if the notion of ``color class'' aggregations is conceivable under a distance definition.

Illuminants

You will use image1 and its other aspects under different illuminants image2, image3, and image4.

The aim here is to see the differences between color rendering and color discrimination from global and local aspects.

About this document ...

Scenes Understanding and Color Spaces

This document was generated using the LaTeX2HTML translator Version 99.2beta8 (1.43)

The command line arguments were:
latex2html -split 0 tp.tex

The translation was initiated by Tera Ovaskainen on 2002-07-23

Footnotes

...colorspace ¹: Any comments, suggestions, remarks should be sent to colantoni@vision.univ-st-etienne.fr

Tera Ovaskainen 2002-07-23