WMSCISSORS HELP FILE
======================
code written by Vlad ATANASIU 2006.09.16

1. Purpose of the program

This code takes as input binary scans of a watermark repertory page 
exhibiting a number of handdrawn watermarks and segments the page in 
smaller images according to the evaluated number of watermarks.

2. Algorithm description

The code is first determining the objects with the 'biggest' convex areas,
then attributing all the objects whos convex hull centroids fall in
the watershed of one of the 'big' objects to the respective object.
'Big' objects are identified from the histogram of the object's convex 
areas: all those with higher values from the point on in which the 
curvature of the histogram is maximal. The code takes as minimum four
objects to be 'big'.

3. Usage instructions

3.1 Batching type
When the dialog box ask you...
<Select how many images in a directory to process>... choose...
<Single image> if you want to process just one image;
<All, supervised> for processing all images in a directory
and be able to make modification of the segmentation proposed
by the program;
<All, automatic> to accept the segmentation done by the machine.

3.2 Image selection

In the next dialog box you select the image you want to 
process. If you want to process all the images in a directory,
select one of the images in the respective directory. In 
this way the machine will know which is the intended directory.

Supported image types appear in the drop-down menu of
the dialog box. Note that uncompressed tif images are supported,
but not compressed tif files.

3.3 Objects reclustering

If you selected to process single images or all images, under
supervision, the program will show you an image with the 
proposed segmentation of the input image into several smaller
images, differentiated by color.

If you consider that a specific object (i.e. shape made of 
connected pixels) or object cluster has to belong to another
cluster than that proposed by the program, you can recluster
the objects.

When moving with the mouse over the image you will see 
a cross-hair appear. Click on the object you want to reattribute
to another cluster, then click on the target cluster. The source
object will change its color to that of the newly attributed cluster.
It is not necessary to click exactly on a object's pixel, since the 
program is automatically selecting the nearest pixel to that 
of where you clicked like a magnet would do. An object is 
defined as the totality of contigous pixels.

By default you select entire groups of objects. If you wish
to reattribute just one single object, press the <s> key. 
You go back to group selection by pressing <g>.

If you want to create a new group press <n>, then click on
an object wich will be part of the new group, then attribute
new objects to it.

You undo the last rattribution by pressing <z>. When you 
finished reclustering press the <space> bar.

3.4 Continue processing

You can continue processing images by clicking the right
button of the message box asking you if you want to continue.


