Conference hydra::amiga_v1

    Ok, I answered my own question.  I simply save the image using the
    Sculpt saver when it's a b&w image.  Then compose the three using the
    Sculpt loader.
    
    The reason I didn't think of this right off the bat is because I'm used
    to TAD, and TAD doesn't have the Sculpt saver included, so...
    
    But I'd like to see this note keep running on the topic of image
    processing.
    
    My next question is:  Are there any good books or people who know etc.
    how a convolution matrix works on a picture?  There are the preset ones
    in the manual, but they don't explain what the numbers do.  Can someone
    help me figure out the math behind making a convolve matrix?
    
    I'm very good at math, by the way, I've been through Calc III and
    Linear Algebra I&II (although II is pretty shaky; I still have the
    books though).
    
    Again, any help would be greatly appreciated.
    
    Mike

There is an image processing program free on a Fish disk, I have it, called
ImageLab on disk 452.

The van dam graphics bible probably has examples of convolution in it.
I think that the matrix just multiplies corresponding elements around the
pixel and adds them up, but I'm unclear.  I've written s/w for convolution
and deconvolution on 'scope waveforms in one dimension, and that's how
it works there.

the asterisks are pixels in the image.  The | are boundaries showing where
the matrix (only 3*3 here) is applied.  The corresponding elements of the
matrix and images are multiplied and summed and put into the pixel in the
center.  At least I think so.  This is done for every element.  The matrix
can be as big as you like, but it's easier if it's square and odd sized.


* * * * * * * * * * * * * * * * * * * * * *
* * * * * * *|* * *|* * * * * * * * * * * * 
* * * * * * *|* * *|* * * * * * * * * * * *
* * * * * * *|* * *|* * * * * * * * * * * *
* * * * * * * * * * * * * * * 


Laplacians are probably a quotient of adjacent pixels to find a rate of
change.  A differencing would be close enough.

Tom
wild guesser

The manual imagelab.doc with imagelab on fish 452 describes all the different
matrices for convolution, I was close.
I have it now and am building a sepaerate imagelab lharc file for myself.
Tom

I've seen a couple of references to "Convolution Matrices" in adpro, but
no description of what the feature really is/does. Could someone give
a description of that feature? If it's as powerfull as the software in
"The Digital Darkroom" (which I tried to talk ASDG into adding to ADPro),
then that's an important addition to ADPro...

	Thanx,
	<mike

    RE: -.1
    
    Right, Mike, that's what I was asking about in .1.  I'm figuring out
    some stuff (will download ff452) but from what I can see the matrix is
    just dot producted across a set of nine pixels, and then that value is
    divided by the ABS of the matrix (to account for linear scaling).  I
    have cross posted this note in Imaging, so if I get a real clear
    cookbook formula for all the equations underneath, I'll post 'em here.
    
    Mike
    

	In terms of the image, convolution ordinary smooths out the image.
	It can act to anti-alias an image.  It blurs lines.  This matrix can
	just have all equal=1 members.  Oscilloscopes that are samplers and
	have smoothing functions work like that in one dimension to
	disguise noise.
	However, special values in the
	matrix can also perform other functions.  A Laplacian matrix
	(negative numbers in pixels around the pixel of interest and positive
	in the pixel balanced to subtract weighted values for surrounding
	values), acts like solarization, leaving only an outline.  Using
	a laplacian matrix a little out of balance in favor
	of the pixel being processed can just sharpen an image.
	Using the same matrix overbalanced for the neighboring pixels
	highlights the big flat areas.
	You can make this matrices assymetrical and make them act only	
	horizontally or vertically.

tom

    I don't know a thing about how convolution works, but in avid it gave
    some examples of what it could be used for. 
    	Convolution matrices- sharpen images
    	Deep press- make you pic appear as if it was pressed into a deep
    	material like leather.
    Avid has a very article on ADPro 2.0 this month.
    							bill

MathVISION from Seven Seas Software (originally Doug's Math Aquarium) has some 
hooks available for setting up your own convolution matrices.

FWIW,

Jim

    Check out Black Belt Systems' Image Processor.  It will do sep RGB
    files, merge, etc!  Very powerful, and virtually free!  Call their BBS
    or you can get it from CompUServe, etc.
    
    Clark