Revision as of 08:06, 9 April 2013 by Mboutin (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)


Lab 7 Discussion, ECE637, Spring 2008

Discussions on Lab 7 goes here.

error diffusion makes image brighter --karl.ostmo.1, Wed, 19 Mar 2008 15:03:39 -0400 reply

I'm pretty sure I followed the error diffusion algorithm and equations given in the lab to the letter, but my image comes out a lot brighter than the original. The error-diffused result looks pretty nice (I would say better than the ordered dithering result), but I did a quick average of the pixel brightness: sum(sum(X))/prod(size(X)), and the original image is ~108.5 while my error-diffused image is ~164.9. Any thoughts?

error diffusion makes image brighter --karl.ostmo.1, Wed, 19 Mar 2008 16:30:26 -0400 reply

I found out what the problem was: imread('house.tif') produces a matrix of uint8 type. This can cause problems when you are distributing the error, if your arithmetic is done in a certain way. MATLAB has counterintuitive behavior when dealing with "int" types. Observe the following MATLAB command sequence:

>> double(1.0/2.0) * 1 ans =

   0.5000

>> double(1.0/2.0) * uint8(1) ans =

   1

Recall that in C, if any term in the expression is a "float" type, the float type dominates the expression, such that ((int) 1)/((float) 2) produces the correct answer of 0.5. However, it appears that in MATLAB, the "uint8" type dominates if it is found anywhere in the expression, which leads to uint8(1)/double(2) producing 1. MATLAB also "rounds" integer division instead of performing the "floor" function::

   >> double(1.0/3.0) * uint8(1) ans =
       0
   >> double(2.0/3.0) * uint8(1) ans =
       1

Alumni Liaison

Basic linear algebra uncovers and clarifies very important geometry and algebra.

Dr. Paul Garrett