Click here to Skip to main content
65,938 articles
CodeProject is changing. Read more.
Articles / Languages / VisualC++

Fast 2:1 Image Shrink (Scale Down)

4.93/5 (24 votes)
15 Aug 2024CPOL3 min read 45K   1.6K  
Fast implementation of 2:1 image shrink (full image or part of it)
This article includes a fast implementation of 2:1 image shrink algorithm.

Image 1

Introduction

This article offers a simple and fast, practical implementation of 2:1 image shrink algorithm.
2:1 image shrink is used in very specific cases, but it does a nice job.

Background

Image processing is always an expensive job. When one needs to show a scaled picture, it is always better to use GPU for rendering (OpenGL, DirectX). But using GPU is not always possible and it is good to have an alternative.

The high-quality image resize (enlarge/shrink) is extremely CPU demanding, especially on large images. In addition, sometimes it is required to do the processing multiple times. To relieve this, there are different tricks that could be applied. If a picture has to be shrunk (scaled down), a trick that is similar to mip-maps could be applied. Such case is if we want to shrink a picture more than twice (destination size is less than half of the original), we could speed up the whole process by applying half-shrink prior to actual scale. 2:1 scale-down is very fast, due to simplest algorithm that uses less computations.

For example, if we have image 1920x1024 and need to shrink it to 500x281, we have to execute code like this:

C++
shrink( imageDst, imageSrc )

But we could optimize this by pre-shrink 2:1 like this:

C++
shrinkHalf( imageHalf, imageSrc );
shrink( imaheDst, imageHalf );

In one of my projects, when I had to scale the image of 3 monitors to generate a preview, the result was:

Scale: 46 ms
ShrinkHalf + Scale: 25 ms
ShrinkHalf + ShrinkHalf + Scale: 16 ms

Which means, the trick applied once, causes %36 speedup.
And the trick applied twice, causes %66 speedup.

The demo application has two buttons:

  • Shrink 2:1 - performs image shrink of the whole source image
  • Shrink 2:1 rnd - performs image shrink of random rectangle from the source image

On the left, there is the original image. On the right, there is the original image in grayscale and the shrunk image (or shrunk part) in color.

Shrinking only parts of the image is appropriate when the original was updated and there is a need to update the shrunk copy too. This could save a lot of CPU and time for processing.

2:1 of the Whole Picture

Image 2

2:1 of Random Rectangle

Image 3

Using the Code

There is a pair of files in the project: ShrinkHalf.h and ShrinkHalf.cpp
To use the code, you only need to put these files into your project and include the ShrinkHalf.h file.
These files contain the implementations of all algorithms.

shinkHalf function has the following arguments:

  • BYTE* dstPixels - destination array of pixels in format 24 bit BGR (b,g,r, b,g,r, ..., b,g,r)
  • BYTE* srcPixels - source array of pixels in format 24 bit BGR (b,g,r, b,g,r, ..., b,g,r)
  • int srcWidth - width in pixels of the image
  • int srcHeight - height in pixels of the image

So, in order to use the functions, you should do the following:

C++
#include "ShrinkHalf.h"

...

// This will perform 2:1 scale-down with color averaging
shrinkHalf( dstPixels, srcPixels, srcWidth, srcHeight );

Note that, the destination does not provide dimensions, as they are calculated from the source sizes.

Here are the functions available:

C++
/////////////////////////////////////////////////////////////////////////////
//    Fast and smooth 2:1 image resize
/////////////////////////////////////////////////////////////////////////////

void    shrinkHalf    ( BYTE* target, const BYTE* source, int srcWidth, int srcHeight );
void    shrinkHalfPart( BYTE* target, const BYTE* source, int srcWidth, 
                        int srcHeight, int x1, int y1, int x2, int y2 );

/////////////////////////////////////////////////////////////////////////////

The first function performs 2:1 shrink of the whole image.
The second function performs 2:1 shrink of the specified rectangle from the source image, calculating the respective rectangle for the destination image.
If you use this, you should pay attention that you could need a correction of +1/-1 of the source rectangle points, because of the integer roundup.
This is very useful if there are intensive updates that should be applied (for example, video rendering).

Points of Interest

In the shrinkHalf function, there is a commented code that works, but I preferred to leave the internal call to shrinkHalfPart.

C++
//static
void    shrinkHalf( BYTE* target, const BYTE* source, int srcWidth, int srcHeight )
{
    shrinkHalfPart( target, source, srcWidth, 
                    srcHeight, 0, 0, srcWidth-1, srcHeight-1 );

    //~~~~~
/*
    int    dstWidth    = srcWidth  / 2;
    int    dstHeight   = srcHeight / 2;

    int    srcLineBytes    = srcWidth * 3;
    int    dstLineBytes    = dstWidth * 3;
    int    dstRows         = dstHeight;

    const BYTE*    sl    = source;
          BYTE*    tl    = target;
    const BYTE*    te    = target + srcHeight/2 * dstLineBytes;

    while( tl < te )
    {
              BYTE*    pt    = tl;
        const BYTE*    p1    = sl;
        const BYTE*    p2    = p1 + srcLineBytes;
        const BYTE*    pe    = sl + srcLineBytes;

        while( p1 < pe )
        {
            *pt++    = (p1[0] + p1[3] + p2[0] + p2[3]) >> 2;    //    / 4;    blue;
            *pt++    = (p1[1] + p1[4] + p2[1] + p2[4]) >> 2;    //    / 4;    green;
            *pt++    = (p1[2] + p1[5] + p2[2] + p2[5]) >> 2;    //    / 4;    red;

            p1    += 6;//2*3;
            p2    += 6;//2*3;
        }

        sl    += srcLineBytes << 1;    //    Shift by 1 is equal to " * 2 "
        tl    += dstLineBytes;
    }
*/
}

So, if someone wants a clean and simpler function, thou could remove the first line that calls shrinkHalfPart and uncomment the commented code.

History

  • 8th April, 2020: Initial version

License

This article, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)