Forums

Image Processing and Human Perception of Distortion

Started by Unknown November 21, 2003
I have a very fundamental question about image processing and
distortion.

Suppose that we have two images: I[i,j] and J[i,j], where 
"i" ranges from 0 to 1919 and "j" ranges from 0 to 1199.
Suppose that we create J[] from I[] by adding uniform white
noise at each pixel of the image.  Assume that each pixel is
24-bit RGB and that we add uniform white noise, independently,
to the red plane, the green plane, and the blue plane.

Suppose that MSEr is the mean squared error between the red
planes of I[] and J[].  Similarly, suppose that MSEg and MSEb
are the mean squared errors between the green planes and
between the blue planes, respectively.

We assume that MSEr = MSEg = MSEb.

Now, here is the question concerning human perception.

What is the maximum value of MSEr such that most people cannot
detect a difference between I[] and J[].

Please respond publicly and by e-mail.
lemon@arithmetic.stanford.edu wrote:

> What is the maximum value of MSEr such that most people cannot > detect a difference between I[] and J[].
Maybe it depends on the APL (Average Picture Level). bye, -- Piergiorgio Sartor