Alpha-Aware Resize

[Reptorian1125]

One of my next attempt at touching CImg,h is implementing alpha-aware resize. That's doable in G'MIC, but the code to do that is resize using average interpolation, split the alpha from color, scale alpha from 0-255 to 0-1, divide color by alpha, re-scale the alpha, append them together. That could be a command without having to do this, but I wanted to do a new command via CImg.h and editing gmic.cpp(?) just to trim some time off (See commentary below) and I'd like that to complete my Mosaic refactor. I can't say I'm satisfied with the execution time via my own G'MIC implementation yet, and I'm thinking it can be noticeably faster.

So, my goal is to have a built-in command which isn't a new function that enables all of that at once.

At the moment, I am studying lines close to this:

//! Resize image to new dimensions \newinstance.

Just so that I can understand it and then make my own implementation.

The plan is to first loop over all of the alpha-channel. Find the weighing factor. And then the color is calculated by multiplying the color by weighing factor of alpha multiplied by alpha. Then insert the result in the resized image. I do wonder if it'll be faster to have a C++ code that stores positions, and amount of alpha value, then use those to find the new value of color which was the main reason I was thinking to touch Cimg.h.

You can see a bit of math here.

# C = Color
# NC = New Color Value
# A = Alpha
# NA = New Alpha Value
# SA= Sum of Alpha
# WF = Weighing Factor

C = 85,255,55
A = 5,100,150

SA =255
WF = 1 / SA = 0.0038461538461538464
NA = 255 / 3 = 85

# To calculate NC, for each color, multiply the value of color by the corresponding alpha value. Multiply the resulting value by the weighing factor. Then, add them altogether

NC = sum(85*5*WF, 255*100*WF, 55*150*WF) =>134.01960784313727

So therefore, the new pixel value is [ 134.01960784313727 , 85 ] .

[dtschump]

Hello @Reptorian1125 ,

I have two things to say about this:

• It seems that the resize algorithm you want to apply is just resizing the alpha-premultiplied image. For this in G'MIC you have commands premula and ipremula which do the job of converting from RGBA to premultiplied RGBA (and the reverse). Just put a resize between and it should do the job.
• I strongly advise against modifying the CImg library just to add very specific features like this in G'MIC. Because CImg is an independent generic image processing library, which is a header file only, and used by other people and for other projects than G'MIC. Then, it's really not at all the intention to add features that will only be of use to a very few people (basically you, the G'MIC user, with your very specific needs).

That's being said, of course, you can have fun figuring out how to add native things to CImg/G'MIC if you like, even make your own forks of CImg/G'MIC, which is very instructive. But there's almost no chance of this kind of functionality being integrated into the official CImg/G'MIC packages, especially if you can do the same thing with a custom G'MIC command.