Skip navigation
Currently Being Moderated

Error in resizing

Nov 24, 2012 6:35 AM

I've found that photoshop is distorting the image when resizing. If you have an image and you duplicate the layer, then with the bottom layer set its size (by free transform) to 99% (width and height), then set the top layers opacity to 50%, the image should be evenly blurred but it's not. It is in focus in the middle and blurred around the edges. Does anybody know why this is happening?

 

222design

 
Replies
  • Currently Being Moderated
    Nov 24, 2012 7:25 AM   in reply to 222design

    Have you actually verified this by creating a difference image? Viewed at 100% zoom and resolution? I think you are merely mistaking some oddities with PS internal tiling system. At certain zoom levels they sometimes do not align perfectly and produce a visible pattern and likewise, some of the tiles may appear less pristine than others. None of that will/ should occur at 100% view. The only other possible influence would be graphics card issues/ limitations where lower res tiles are used to save resources...

     

    Mylenium

     
    |
    Mark as:
  • JJMack
    5,979 posts
    Jan 9, 2006
    Currently Being Moderated
    Nov 24, 2012 7:51 AM   in reply to 222design

    222design wrote:

     

    I've found that photoshop is distorting the image when resizing. If you have an image and you duplicate the layer, then with the bottom layer set its size (by free transform) to 99% (width and height), then set the top layers opacity to 50%, the image should be evenly blurred but it's not. It is in focus in the middle and blurred around the edges. Does anybody know why this is happening?

     

    222design

    No it should not be blurred evenly.  Visulize it this way when transform 99% width and 99% height and anchor point center. the center pixel will not move in other the just resize in where outter pixels will be moved in as will as resized. Center will not blur much outer will. A linner blur from the center.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 8:40 AM   in reply to 222design

    Two months ago, you posted the same thing and it was explained then that the result of the resizing is as should be expected and is correct.

     

    http://forums.adobe.com/message/4787168#4787168

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 8:46 AM   in reply to conroy

    The following was posted last time you reported the same non-error...

     

    If you look at the 99% size image in isolation, you'll see that it has uniformly distributed slight blurring due to resampling.

     

    When it's overlaid at 50% opacity on the original (or vice versa), you are seeing an additional blur due to features in the two images being offset. Obviously, the offset, and therefore its blurring effect, linearly increases from zero pixels at the centre to several pixels (depending on image size) at each corner.

     

    Here's a 100% size original followed by the original overlaid with a 90% reduced duplicate at 50% opacity. You see the offset increasing as the distance from the centre increases.

     

    Screen shot 2012-10-19 at 20.32.59.png

     

    Screen shot 2012-10-19 at 20.33.05.png

     
    |
    Mark as:
  • JJMack
    5,979 posts
    Jan 9, 2006
    Currently Being Moderated
    Nov 24, 2012 9:56 AM   in reply to 222design

    222design wrote:

     

    It should be blurred evenly. The area that the pixel covers should have an equal offset whether it is in the centre or on the outside.

    Lets try again.  Try this approach. start by not moving any pixel center and resize every pixel to 99% of its size. You wind up with an image with gaps between pixels. 

     

    Now lets eliminate these gaps.  Start from the center. Move the pixels around the center pixel in one gap distance. next move out and move the surrounding pixels in two gaps repeat move in three gaps etc...

     

    This is how Image size works when RESAMPLE is not checked and you increase the image's DPI resolution.  Print pixels are smaller the number of pixels remain the same and the print size is reduced in size.  The higher resolution increases the image sharpness.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 9:58 AM   in reply to 222design

    It doesn't matter whether the pixel is at the centre or the edge the offset will be the same,

     

    No, it won't. Scaling is not a linear operation. The discrepancies in resampling increase the farther away a pixel is from the transform pivot and at large settings, this will cause notable visual degeneration. It should, however, not result in extreme blurriness with normal scaling operations like upsizing by 150% or scaling down to 99%.

     

    Mylenium

     
    |
    Mark as:
  • JJMack
    5,979 posts
    Jan 9, 2006
    Currently Being Moderated
    Nov 24, 2012 10:27 AM   in reply to Mylenium

    Scaling an Image width and height by the same percentage should be liner be it done by changing pixel size or interpolating the number of pixels in the image.  When resampling an image's image quality will suffer for the new pixels are computer generated using some interpolation method. Not captured with optics and a sensor.  Pixel quality of a resampled image will not be as good as the pixel quality of the original image.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 10:53 AM   in reply to 222design

    222design wrote:

     

    Yes but this is incorrect. When you change the layer size to 99% it resamples the image producing a different grid of area covered by a pixel. It doesn't matter whether the pixel is at the centre or the edge the offset will be the same, so the image should be evenly blurred.

     

    Your thinking is incorrect.

     

    The offset must increase as the distance from the scaling origin is increased. There is no possible other outcome. The blur therefore will increase with distance from point of alignment of the stacked images.

     

    Consider the scaling in one dimension to simplify matters.

     

    Here's an original row of squares and a copy at 100% scale.

     

    100 percent.png

     

    Now the copy row has been scaled to 99%. See the offset increase from left to right.

     

    99 percent.png

     

    Finally, when overlaid, there is the explanation for the blur not being contant across your stacked layers.

     

    overlaid.png

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 11:20 AM   in reply to 222design

    The images are demonstrating what you saw in your own stacked image, yes, but there is no error. You'll help your argument that there is an error in Photoshop's scaling if you post an example of "correct" scaling which conflicts with Photoshop's scaling.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 11:37 AM   in reply to 222design

    222design wrote:

     

    how can I post an example of correct scaling with photoshop when photoshop has the scaling error in it?

     

    You make a mock-up of the result which you consider correct.

     

     

    There is an error in your images as they are done with photoshop and photoshop has this error in it.

     

    You have yet to demonstrate that.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 11:42 AM   in reply to 222design

    You surely know your own expectation of my row of squares when resized to 99%. That can't be difficult to illustrate.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 12:22 PM   in reply to conroy

    conroy wrote:

     

    You surely know your own expectation of my row of squares when resized to 99%. That can't be difficult to illustrate.

    Ok, I'll illustrate what 222design expects

     

    the black lines represent the pixel grid in Photoshop. The blue squares fitting in each pixel represents the pixels of the original image. The small red squares is 222design's expectation of scaled image that will produce an even blur.

     

    Untitled-8.jpg

     

    @ 222design, the problem with your expectation is that it is not possible for a pixel to be smaller than a cell of the pixel grid, and as explained many times before there will be always one center of scale for all pixels. It is not possible to scale each pixel individually from its own center.

     

    so, the following statement is incorrect:

    222design wrote:

     

    ....When you change the layer size to 99% it resamples the image producing a different grid of area covered by a pixel. ..

    the pixel grid never changes, it is absolute.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 12:42 PM   in reply to 222design

    222design wrote:

     

    yes I know that the pixel grid never changes and as I have stated what does change is the area covered by the pixel and this should be a constant offset giving a constant blur....

    And because the grid never changes but only the area covered by the scaled pixels, the only possible ways for this to happen are illustrated on the images below. The blue pixels show where the burring will occur and cannot be the same for all pixels.

    Untitled-8.jpg

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 1:26 PM   in reply to 222design

    >> this is independent of position of pixel, it is linear

     

    That is the part you keep getting wrong.

    The scaling is very much DEPENDENT on the position of the pixel.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 1:26 PM   in reply to 222design

    >> this should be a constant offset

     

    No, that would be the effect if you offset your layer by a fraction of a pixel.

    Scaling will result in a linear change xNew = offset + scale * xOld

    So pixels near the center will have smaller offsets, and those near the edge have larger offsets.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 1:47 PM   in reply to 222design

    Um, no. That is not how it works.

    Please try the examples given, or try it on some graph paper.

    It cannot be uniform because you are working on a fixed grid.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 2:07 PM   in reply to 222design

    222design wrote:

     

    the resampling is the same at the edges and at the centre it is linear or uniform so it should be constantly blurred.

     

     

    There are two sources of blur in your example of an overlaid image of different scale and 50% opacity. Maybe you missed me telling you that last month and earlier today:

     

    "If you look at the 99% size image in isolation, you'll see that it has uniformly distributed slight blurring due to resampling.

     

    When it's overlaid at 50% opacity on the original (or vice versa), you are seeing an additional blur due to features in the two images being offset. Obviously, the offset, and therefore its blurring effect, linearly increases from zero pixels at the centre to several pixels (depending on image size) at each corner."

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 2:45 PM   in reply to 222design

    222design wrote:

     

    but the resampling to 99% is linear, it does not matter whether it is an outer or inner pixel, so the offset is constant, so it should be uniformly blurred.

     

    The main blur that you are seeing, the one that increases toward the outside of the image, is not a resampling blur, though. It is a perceptual blur due to offset of features in the differently scaled stacked images. Even if the images were composed of an infinite number of infinitely small pixels and the resized one had no resampling blur, the offset due to scaling would still result in you perceiving a blur that increases from zero at the centre to maximum at the corners.

     
    |
    Mark as:
  • Currently Being Moderated
    Nov 24, 2012 2:59 PM   in reply to 222design

    PLEASE go try this on graph paper.

    It is linear, so you will NOT get uniform blurring/antialiasing on a finite grid.

     
    |
    Mark as:

More Like This

  • Retrieving data ...

Bookmarked By (0)

Answers + Points = Status

  • 10 points awarded for Correct Answers
  • 5 points awarded for Helpful Answers
  • 10,000+ points
  • 1,001-10,000 points
  • 501-1,000 points
  • 5-500 points