Announcement

Collapse
No announcement yet.

Question about "Resize (faster, lower quality)" algorithm

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

    Question about "Resize (faster, lower quality)" algorithm

    Hello, first time poster here!

    Does anyone know which resize algorithm is implemented for the "Resize (faster, lower quality)" checkbox in the "image->Resize/Resample" menu option?

    I'm doing a project at school which involves downscaling of high resolution circuit diagrams with lots of fine straight lines and sharp edges. The relevant color depth is 1-bit and 4-bit. Irfanview's "Resize (faster, lower quality)" works very well when scaling down to even 3-5% of original resolution.

    See the attached picture for the exact option I'm referring to.

    Thanks in advance!

    Click image for larger version

Name:	irfan-resize.png
Views:	1
Size:	26.2 KB
ID:	85299
    Last edited by lessismore; 11.04.2020, 01:08 PM. Reason: added a quesiton mark to the question

    #2
    If I remember rightly it uses an algorithm written by the program's author, Irfan Skiljan.

    Can't you save your drawings in a vector format to avoid scaling problems? SVG or PDF might be best.
    Before you post ... Edit your profile • IrfanView 4.62 • Windows 10 Home 19045.2486

    Irfan PaintIrfan View HelpIrfanPaint HelpRiot.dllMore SkinsFastStone CaptureUploads

    Comment


      #3
      Our input is pointer to a mmaped bitmap image of very high resolution. Output must be downscaled bitmap.

      I dont have any flexibility in those matters.

      However it seems the fast/naive algorithms work really well with this particular type of input, whereas Lanczos, Mitchell, Bspline and other more expensive algorithms really dont do well at all. The images are mostly b&w with a few colors used here and there (sparingly).

      I suppose it is some kind of nearest neighbour type algorithm. It seems the simple algorithms in GIMP also do really well.

      Comment


        #4
        It looks like point sampling (nearest neighbor). Don't you encounter fine lines that disappear entirely from the diagram when skipped by the simple filter?

        A weighted average ("fast") should be better in all cases, followed by a gamma correction to make the lines darker and requantization if needed. Lanczos or cubic should work on a white background where you can't see the overshoot, and anti-alias curved lines better.

        Comment

        Working...
        X