In Quartz 2D, is it possible to mask an image by removing everything but the color channel you want?

Hmmm ...

Maybe I am missing something, but I do not believe the provided answers apply to the question. The second answer comes close to the sign, but contains a false code that has nothing to do with the solution.

The method createMask

makes a copy of the original image, assuming alpha at the LSB position. ChangeColor makes a masking call that won't affect the RGB image much - basically, only black will be masked (i.e. RGB triplets in the range / combinations (1,1,2) to (3, 2,1)).

I assume that the observed red shift is due to the parameter

kCGImageAlphaPremultipliedFirst

in line

CGContextRef cgctx = CGBitmapContextCreate (bufferdata,mWidth,mHeight, 8,mWidth*4, colorSpaceRef, kCGImageAlphaPremultipliedFirst);

      

        
        
        
      

    

due to improper alpha channel treatment. If in the method changeColor

you change the block

CGImageRef ref1=[self createMask:temp23];
const float colorMasking[6] = {1.0, 3.0, 1.0, 2.0, 2.0, 3.0};
CGImageRef New=CGImageCreateWithMaskingColors(ref1, colorMasking);
UIImage *resultedimage=[UIImage imageWithCGImage:New];
EditImageView.image = resultedimage;

      

        
        
        
      

    

to be

CGImageRef ref1=[self createMask:temp23];
UIImage *resultedimage=[UIImage imageWithCGImage:ref];
EditImageView.image = resultedimage;

      

        
        
        
      

    

you will not see any difference in display. Changing the CGBitmapInfo constant to kCGImageAlphaPremultipliedLast

display the image correctly using any of the specified code blocks.

The following answer comes a little closer to what the OP requires, but it's in terms of visual effect, not actual data. Here the corresponding code in createMask

is

CGContextRef cgctx = CGBitmapContextCreate (bufferdata,mWidth,mHeight, 8,mWidth*4,colorSpaceRef, kCGImageAlphaPremultipliedLast);

      

        
        
        
      

    

which displays the image correctly and then

CGContextSetBlendMode(cgctx, kCGBlendModeColor); 
CGContextSetRGBFillColor (cgctx, 1.0, 0.0, 0.0, 1.0);
CGContextFillRect(cgctx, rect);

      

        
        
        
      

    

and then the logic for building the image. The blending logic applies a red tint to the original image, achieving the same effect as the disordered alpha channel in the original response. This is still not what the OP is asking for, which should mask one or more channels rather than blend colors.

This really means setting channel values ​​for colors that are not desirable for zero. Here's an example of only returning the red channel as OP requests; the pixel format is assumed to be ABGR:

- (CGImageRef) redChannel:(CGImageRef)image 
{
    CGDataProviderRef provider = CGImageGetDataProvider(image);
    NSMutableData* data = (id)CGDataProviderCopyData(provider);

    int width = CGImageGetWidth(image);
    int height = CGImageGetHeight(image);

    [data autorelease];

    // get a mutable reference to the image data
    uint32_t* dwords = [data mutableBytes];

    for (size_t idx = 0; idx < width*height; idx++) {
        uint32_t* pixel = &dwords[idx];
        // perform a logical AND of the pixel with a mask that zeroes out green and blue pixels
        *pixel &= 0x000000ff;
    }

    CGColorSpaceRef colorSpaceRef = CGColorSpaceCreateDeviceRGB();
    CGBitmapInfo bitmapInfo = kCGBitmapByteOrderDefault;
    CGColorRenderingIntent renderingIntent = kCGRenderingIntentDefault;

    // now create a new image using the masked original data
    CGDataProviderRef iprovider = CGDataProviderCreateWithData(NULL, dwords, width*height*4, NULL);
    CGImageRef savedimageref = CGImageCreate(width, height, 8, 32, width*4, colorSpaceRef, bitmapInfo, iprovider, NULL, NO, renderingIntent);
    CGColorSpaceRelease(colorSpaceRef);
    CGDataProviderRelease(iprovider);

    return savedimageref;
}

      

        
        
        
      

    

A good summary of bitwise operations can be found here .

As indicated here , you may need to change the mask structure based on the LSB / MSB order of the bits in the pixel. This example assumes 32-bit pixels from a standard iPhone PNG.