How to perfectly fit any image on a fixed size picture control

[jeeswg]

@SKAN: Where did you get the maths for the ScaleRect function? Btw is Floor necessary, or can Round be used instead? Because presumably Round would give a marginally closer aspect ratio in some cases.

I've found this quite a tricky issue. Here's my current function (which uses Round):

Code: Select all

;shrink image to fit rectangle (maintain aspect ratio)
JEE_MathsImageFitRectShrink(vImgW, vImgH, vLimW, vLimH, ByRef vImgW3, ByRef vImgH3)
{
	vImgW3 := vImgH3 := ""
	if (vImgW <= vLimW) && (vImgH <= vLimH)
	{
		vImgW3 := vImgW, vImgH3 := vImgH
		return
	}
	vRatioW := vLimW / vImgW ;to fit the image width-wise (although might be too high)
	vRatioH := vLimH / vImgH ;to fit image height-wise (although might be too wide)
	vImgH1 := Round(vImgH * vRatioW) ;vImgW1 := vLimW
	if (vImgH1 <= vLimH)
	{
		vImgW3 := vLimW, vImgH3 := vImgH1
		if (vRatioW > vRatioH)
			return
	}
	vImgW2 := Round(vImgW * vRatioH) ;vImgH2 := vLimH
	if (vImgW2 <= vLimW)
		vImgW3 := vImgW2, vImgH3 := vLimH
	;diagnostic:
	if (vImgW3 = "")
		MsgBox, % "error: " A_ThisFunc "`r`n" "img: " vImgW " " vImgH "`r`n" "lim: " vLimW " " vLimH "`r`n" vImgW1 " " vImgH1 "`r`n" vImgW2 " " vImgH2
}

I also wrote a script to test it:

Code: Select all

q:: ;confirm that JEE_MathsImageFitRectShrink never returns blanks
vOutput := ""
vNum := 51
Loop, % vNum
{
	vLimH := A_Index-1
	Loop, % vNum
	{
		vLimW := A_Index-1
		Loop, % vNum
		{
			vImgH := A_Index-1
			Loop, % vNum
			{
				vImgW := A_Index-1
				;if the function fails, then:
				;a MsgBox within the function should trigger
				JEE_MathsImageFitRectShrink(vImgW, vImgH, vLimW, vLimH, vImgW3, vImgH3)
			}
		}
	}
}
Clipboard := vOutput
MsgBox, % "done"
return

[SKAN]

@SKAN: Where did you get the maths for the ScaleRect function?

I took some help.. 9 years ago from [VxE]
https://autohotkey.com/board/topic/2782 ... ixed-rect/
But there were rounding errors in those functions.

Btw is Floor necessary, or can Round be used instead? Because presumably Round would give a marginally closer aspect ratio in some cases.

The image has to be exact fit at least on one side of the target rect.
So I return one value as is ( either TW or TH ) and only compute the other based on the aspect ratio.
In tricky situations, eg. Source = 256x257 and target = 257x256 I wouldn't want round upwards, hence I went with floor.
OMG! I didn't test the above.. I forgot.

Thanks for sharing your version. I will test it soon.

[SKAN]

I don't think it breaks anything

I think semi transparent pixels are lost... I'm quoting from memory.

[Drugwash]

… and I was quoting from… hopes.
Frankly I have no idea. Do you have some specific alternative API set and image to compare the results?

[EDIT]
Unfortunately there is something going on and it's not for the best. I just tested the script with the AHK logo and apparently the blur effect doesn't catch on. it works with the 24bit PNG poster but not with the 32bit alpha-channel PNG logo. Tested both in 98SE and XP (just to be clear).

[jeeswg]

I think that in my example, it is OK to use Round (marginally more accurate maintenance of ratio), and not Floor:

Note: we try two possible pairs of values:
W * (LimW/W) = LimW and H * (LimW/W)
W * (LimH/H) and H * (LimH/H) = LimH

start: W, H [both above their limits][where LimW/W > LimH/H]
limits: LimW, LimH
[note: if 'try 1' values too big to fit rectangle, try 'try 2' values]
try 1: LimW, Round(H*(LimW/W)) [bigger than 'try 2' values]
try 2: Round(W*(LimH/H)), LimH [smaller than 'try 1' values]
is Round(W*(LimH/H)) definitely <= LimW? (in which case at least one 'try' would always work)

since W*(LimH/H) <= LimW
so, Round(W*(LimH/H)) <= Round(LimW)
since LimW is an integer, so Round(LimW) = LimW
and so, Round(W*(LimH/H)) <= Round(LimW) = LimW
so, Round(W*(LimH/H)) <= LimW as required

If someone can prove me wrong, or find relevant info for this problem, or generally just find a better more succinct function than mine, that would be most welcome.

@SKAN: Funnily enough when you posted this script, I was thinking about this very problem.

[tmplinshi]

Works great!! Thank you.

[Jose Hidalgo]

Hi all, thanks for this nice topic. I hope somebody's still listening.
I'm trying to use the method described by @SKAN in this post, but so far only the GDI+ image displays, not the Lanczos one.
I have put Resample.dll, Resample.h and Resample.lib in the same folder as my script. Am I missing something ?



Also, my final purpose would be to include the Resample files in a compiled executable. I could include the files with FileInstall, but is there a better way ? And could they be in a different folder ? (a Resources folder for instance).

[SKAN]

I'm trying to use the method described by @SKAN in this post

You could try my Imagen() to scale image.

[Jose Hidalgo]

OMG Imagen() works beautifully ! So much better than the standard scaling !! Thank you SKAN, I'll be including it right away !

[Sam_]

its a limitation of GDI/GDI+ from what I understand, you have to roll your own
https://www.codeproject.com/Articles/11143/Image-Resizing-outperform-GDI

Thanks for the link.
There is a 32bit DLL based on that code
Plain C Resampling DLL : https://www.codeproject.com/Articles/22271/Plain-C-Resampling-DLL

Wish AHK have this incorporated.. Not sure if CPOL License is compatible with GPL v2

I don't suppose you or any other kind soul would be willing to recompile this into a 64-bit dll? Also, a couple of bug fixes have been posted in the discussion that don't seem to have been implemented yet.

[gwarble]

Have you tried Imagen() from Skan yet? Dll shouldn't be required unless I'm missing something

ahk.ahk-20230322141806.png (16.38 KiB) Viewed 1474 times

Code: Select all

Gui, Margin, 0, 0
;pic is 660x100
Gui, Add, Picture, x0 y0  w165 h25, ahk.png
Gui, Add, Picture, x0 y25 w165 h25 hwndhPic
Imagen("ahk.png","ahk_id" . hPic)
Gui, Show

[Sam_]

@gwarble

As part of a larger project, I want to allow the end-user to be able to select a specific InterpolationMode/resizing algorithm/resampling filter to be used to resize a set of image frames in an animated graphic. I've already implemented the ones GDI+ supports, but thought this dll had some interesting and potentially improved options over what GDI+ offers. At a glance it looks like Imagen() only uses the GDI+ InterpolationMode=HighQualityBicubic and SmoothingMode=HighQuality. I'm looking for some additional options. If I can get these implemented via x86 and x64 DLLs, my next task is to look into implementing pixel art scalers, either via DLL if I can find them or manually in AHK if I can't.

[gwarble]

oh ok my bad, you quoted my response to better downscaling specifically, which Imagen() solves nicely for me, but if you're after all the specific modes available then that dll or dissecting the GDI+ code and adding options for other interpolation and smoothing modes might be your best bet, good luck

note to self: https://photosauce.net/blog/post/image-scaling-with-gdi-part-4-examining-the-interpolationmode-values

[Sam_]

note to self: https://photosauce.net/blog/post/image-scaling-with-gdi-part-4-examining-the-interpolationmode-values

That was a very interesting and informative series of blog posts. I learned a lot in the ones I have read so far. Thanks for sharing!