Page 1 of 1

Posted: 2006-11-01T08:23:11-07:00
by magick
Its possible those stripes are not there at all... Try writing to a PNG image. Do you get the same effect? If not, it could be the JPEG lossy DCT transform that is introducing the problem.

Posted: 2006-11-01T17:19:34-07:00
by anthony
The stripes can also be a visual artifact of the display and browser you are using too.
Such stripes have appeared to become more pronounced since i moved to a large LCD display.

Posted: 2006-11-02T16:55:17-07:00
by anthony
Do you have an example of a better gradient image that doesn't show these lines.

I can verify that the gradient function is working as I described in IM Examples.

Hmmm perhaps the lines are caused by the what the gradient not only adds a gradient
vertically, but also along each row. It may be that this is causing a slight moire (interference) pattern with the display.

Try an image like this...
convert -size 1x800 gradient: -rotate -90 -scale 800x600\! test.png

This image ensures the gradient function does not generate any row gradients.

If this is fine as PNG but not as JPEG, let me know.

Posted: 2006-11-05T20:27:34-07:00
by anthony
Nice site. I can see the stripe effect. My thinking is that photoshop uses a color dither, and passes this on in the images it saves. It looks remarkably like what I see on display that does not have true color. greyscale and very drak areas show it much more pominentally.

As Irfanview does not show the stripes in the saved image then it is likely the saved image does not have the striping.

I thought that maybe only JPEg would show the stripes and not the PNG which is a perfect non-lossy algorithm, indicating a problem in the delegate library. However
as you said neither showed striping in Irfanview, but only in photoshop, then the problem is further downstream of the processing.

Remember IM does not actually write JPEG format itself, but uses a delegate library to do this. Prehaps you can try to install the new JPEG2000 library and recompile to see if that makes a difference.
----

What I thought may be happening was something that I recently picked up on by accident while reporting a separte bug in IM (now fixed). On the Bug Testing page of IM Examples
http://www.cit.gu.edu.au/~anthony/graph ... s/testing/
(misplaced as I haven't moved it to its final position) is an example of an effect caused by the lossy algorithim used by JPEG.

This shows how a 4 pixel wide distortion appears between a greyscale and bluescale edge. The best example of the JPEG distortion I have yet seen. PNG does not have this problem. However the strange thing about this example is that the distortion appears reversed. That is the intermedite light and dark band are reversed to what I expected.

However from past experience, this appears to be a true JPEG edge distortion effect, and probably caused by some sort of interference pattern in the fourier transformation involved (wave function effects).

In this case however that does not appear to be JPEG edge effects, but looks to be more like color 'leveling' effects, due to color quantization. both JPEG and PNG should not have any 'levels' effect like this, as these format di not use limited color tables.