[Image-SIG] open, show, png, blank

[Randolph Bentson]

On Thu, Feb 09, 2012 at 12:40:49PM -0500, Chris Mitchell wrote:
> For a linear equation, you would want do use this:
> 
> outJ = outI.point(lambda i:(i-y)*x)

I think this should be outJ = outI.point(lambda i:(i-lo)*x), but the
point conversion method doesn't deal with these lambda definitions.
My definition of y should have been "y = -lo*x" which brings my
code to an algebraic equivalent to what I think Chris intends.

This point conversion is only appropriate for use of the show
method of the new image.  Although outJ.show() creates a properly
formed Netpbm PGM "rawbits" image, if outJ is saved, the resulting
grayscale png file is black because it's a grayscale image which
has been scaled down a lot.

The single statement
 outI.point(lambda i:i*x+y).show()
avoids the temptation to save the rescaled image. :-)

-----------------------------------------------------

Although this design gives a non-white image, it is flawed in that
the displayed image has greater contrast than the original. This
is because the image under discussion doesn't use the full 16 bit
"I" mode range of 0 to 65535. My code maps a lesser source range
to 0 to 255 instead of 32 to 236 which would preserve the contrast.

So I propose a revised solution
  scale = 256.0/2**16
  outI.point(lambda i:i*scale+0).show()

This relies on the "I" mode image having 16 bit pixels. I couldn't
find a way to determine the source pixel size in PIL. The description
of "I" mode in pil-handbook.pdf says it's 32 bit, which suggests
other grayscale files may use other pixel sizes.

-- 
Randolph Bentson
bentson at holmsjoen.com