MUSE im imp++

A part of MUSE (multivariate segmentation) is now available as a plug-in module for imp++. This very brief introduction should help you to get started.

News

9 May 2002: I have removed one limitation; you can now access the "image handler functions" (contrast/brightness, etc) by right-clicking in an image even if the image is currently being edited.

Starting MUSE

I have defined a command impmuse; this means that all you have to do in order to start the program is to enter that command in a terminal window.

If you are running the program over the net you may have

to tell your computer to allow the CBA computer you are using to display its windows on your computer (typically "xhost + xxx.cb.uu.se" where xxx is the CBA computer)
and also to tell the CBS computer where MUSE is supposed to display its windows (typically "setenv DISPLAY xxxxxx" or "export DISPLAY=xxxxxx)" where xxxxxx is the full name of your computer.

Enter xhost on your home computer and setenv/export on the CBA computer.

You need a UNIX computer, or at least an X server running on your computer.

Using MUSE to inspect multispectral images

Select MUSE -> edit 2D in the main menu. This brings up:

Make sure that the correct image (the multispectral one) is selected and press ok (you can ignore inBg). If everything is ok the region editor appears:

In addition to the editing functions this version of the editor has the two buttons Create2D and CreateLayerImage.

CreateLayerImage allows you to create one-layer images for each of the layers in the image. These images are used only for display - you cannot edit regions in them.

If you press Create2D you get a dialog box:

Enter two layer numbers (layers are numbered from 0 so the layers of an RGB image are numbered 0, 1 and 2, for example) and press OK. A 2D histogram image (a.k.a. scatterplot) is computed and displayed

You can edit in either the 2D image or the multispectral image. Select class - or region or whatever you want to call it - by pressing one of the coloured push buttons. It is possible to create several 2D images. When you edit in any 2D image the regions in that image will be mapped to the multispectral image while when you edit in the multispectral images the regions will be mapped to all 2D images.

A snapshot of part of the screen may look like this (the image is Sthlm_Soder and the display shows layers 8, 5 and 0 as an RGB image):

The 2D histogram has been edited so the mapping in this case is from the histogram to the image.

Display of multispectral (more than 3 bands) images

Multispectral images can be displayed either as grey scale (using one band) or RGB (using three bands). Right-clicking on the image and choosing display mode will bring up the following dialog box:

Select RGB or B/W and enter the bands you want to use.

Bugs and limitations...

exist in abundance; this is not a complete list.

Only 8 bit images are supported (I think that the program stops you from using larger pixel sizes but I do not guarantee that it's done in a gentle manner).

The pixels of the 2D histograms are 8 bits but no check for overflow is implemented...

Images

The directory /usr/users/lab/ contains sample images.

Examples:

rgbLena.ipad (RGB image)
Sthlm_Soder.ipad (multispectral images with 14 bands)