Create RGB image from each channel - image

Question

I have 3 files, one with only a red channel, one with only a green channel, one with only a blue channel. Now i want to combine those 3 images to one, where every image is one color-channel in the finished image.
How can i do this with cocoa? I have a solution that is working but is too slow:
NSBitmapImageRep *rRep = [[rImage representations] objectAtIndex: 0];
NSBitmapImageRep *gRep = [[gImage representations] objectAtIndex: 0];
NSBitmapImageRep *bRep = [[bImage representations] objectAtIndex: 0];
NSBitmapImageRep *finalRep = [rRep copy];
for (NSUInteger i = 0; i < [rRep pixelsWide]; i++) {
for (NSUInteger j = 0; j < [rRep pixelsHigh]; j++) {
CGFloat r = [[rRep colorAtX:i y:j] redComponent];
CGFloat g = [[gRep colorAtX:i y:j] greenComponent];
CGFloat b = [[bRep colorAtX:i y:j] blueComponent];
[finalRep setColor:[NSColor colorWithCalibratedRed:r green:g blue:b alpha:1.0] atX:i y:j];
}
}
NSData *data = [finalRep representationUsingType:NSJPEGFileType properties:[NSDictionary dictionaryWithObject:[NSNumber numberWithDouble:0.7] forKey:NSImageCompressionFactor]];
[data writeToURL:[panel URL] atomically:YES];

Answer 1

The Accelerate.framework provides a function to combine 3 planar images into one destination:
vImageConvert_Planar8toRGB888.
I haven't tried your approach but the vImage based method below is quite fast.
I was able to combine three (R,G,B) planes of a 1680x1050 image in ~0.1s on my Mac. The actual conversion takes ~1/3 of that time - The rest is setup & file IO.
- (void)applicationDidFinishLaunching:(NSNotification *)aNotification
{
NSDate* start = [NSDate date];
NSURL* redImageURL = [[NSBundle mainBundle] URLForImageResource:#"red"];
NSURL* greenImageURL = [[NSBundle mainBundle] URLForImageResource:#"green"];
NSURL* blueImageURL = [[NSBundle mainBundle] URLForImageResource:#"blue"];
NSData* redImageData = [self newChannelDataFromImageAtURL:redImageURL];
NSData* greenImageData = [self newChannelDataFromImageAtURL:greenImageURL];
NSData* blueImageData = [self newChannelDataFromImageAtURL:blueImageURL];
//We use our "Red" image to measure the dimensions. We assume that all images & the destination have the same size
CGImageSourceRef imageSource = CGImageSourceCreateWithURL((__bridge CFURLRef)redImageURL, NULL);
NSDictionary* properties = (__bridge NSDictionary*)CGImageSourceCopyPropertiesAtIndex(imageSource, 0, NULL);
CGFloat width = [properties[(id)kCGImagePropertyPixelWidth] doubleValue];
CGFloat height = [properties[(id)kCGImagePropertyPixelHeight] doubleValue];
self.image = [self newImageWithSize:CGSizeMake(width, height) fromRedChannel:redImageData greenChannel:greenImageData blueChannel:blueImageData];
NSLog(#"Combining 3 (R, G, B) planes of size %# took:%fs", NSStringFromSize(CGSizeMake(width, height)), [[NSDate date] timeIntervalSinceDate:start]);
}
- (NSImage*)newImageWithSize:(CGSize)size fromRedChannel:(NSData*)redImageData greenChannel:(NSData*)greenImageData blueChannel:(NSData*)blueImageData
{
vImage_Buffer redBuffer;
redBuffer.data = (void*)redImageData.bytes;
redBuffer.width = size.width;
redBuffer.height = size.height;
redBuffer.rowBytes = [redImageData length]/size.height;
vImage_Buffer greenBuffer;
greenBuffer.data = (void*)greenImageData.bytes;
greenBuffer.width = size.width;
greenBuffer.height = size.height;
greenBuffer.rowBytes = [greenImageData length]/size.height;
vImage_Buffer blueBuffer;
blueBuffer.data = (void*)blueImageData.bytes;
blueBuffer.width = size.width;
blueBuffer.height = size.height;
blueBuffer.rowBytes = [blueImageData length]/size.height;
size_t destinationImageBytesLength = size.width*size.height*3;
const void* destinationImageBytes = valloc(destinationImageBytesLength);
NSData* destinationImageData = [[NSData alloc] initWithBytes:destinationImageBytes length:destinationImageBytesLength];
vImage_Buffer destinationBuffer;
destinationBuffer.data = (void*)destinationImageData.bytes;
destinationBuffer.width = size.width;
destinationBuffer.height = size.height;
destinationBuffer.rowBytes = [destinationImageData length]/size.height;
vImage_Error result = vImageConvert_Planar8toRGB888(&redBuffer, &greenBuffer, &blueBuffer, &destinationBuffer, 0);
NSImage* image = nil;
if(result == kvImageNoError)
{
//TODO: If you need color matching, use an appropriate colorspace here
CGColorSpaceRef colorSpace = CGColorSpaceCreateDeviceRGB();
CGDataProviderRef dataProvider = CGDataProviderCreateWithCFData((__bridge CFDataRef)(destinationImageData));
CGImageRef finalImageRef = CGImageCreate(size.width, size.height, 8, 24, destinationBuffer.rowBytes, colorSpace, kCGBitmapByteOrder32Big|kCGImageAlphaNone, dataProvider, NULL, NO, kCGRenderingIntentDefault);
CGColorSpaceRelease(colorSpace);
CGDataProviderRelease(dataProvider);
image = [[NSImage alloc] initWithCGImage:finalImageRef size:NSMakeSize(size.width, size.height)];
CGImageRelease(finalImageRef);
}
free((void*)destinationImageBytes);
return image;
}
- (NSData*)newChannelDataFromImageAtURL:(NSURL*)imageURL
{
CGImageSourceRef imageSource = CGImageSourceCreateWithURL((__bridge CFURLRef)imageURL, NULL);
if(imageSource == NULL){return NULL;}
CGImageRef image = CGImageSourceCreateImageAtIndex(imageSource, 0, NULL);
CFRelease(imageSource);
if(image == NULL){return NULL;}
CGColorSpaceRef colorSpace = CGImageGetColorSpace(image);
CGFloat width = CGImageGetWidth(image);
CGFloat height = CGImageGetHeight(image);
size_t bytesPerRow = CGImageGetBytesPerRow(image);
CGBitmapInfo bitmapInfo = CGImageGetBitmapInfo(image);
CGContextRef bitmapContext = CGBitmapContextCreate(NULL, width, height, 8, bytesPerRow, colorSpace, bitmapInfo);
NSData* data = NULL;
if(NULL != bitmapContext)
{
CGContextDrawImage(bitmapContext, CGRectMake(0.0, 0.0, width, height), image);
CGImageRef imageRef = CGBitmapContextCreateImage(bitmapContext);
if(NULL != imageRef)
{
data = (NSData*)CFBridgingRelease(CGDataProviderCopyData(CGImageGetDataProvider(imageRef)));
}
CGImageRelease(imageRef);
CGContextRelease(bitmapContext);
}
CGImageRelease(image);
return data;
}

Answer 2

Your program creates many many many many many many color objects.
Although your program could simply access the image reps' bitmapData, it would require your program to know a lot about bitmap representations.
Before taking that approach, you should prefer to let Quartz do the heavy lifting by rendering each image to a CGBitmapContext (e.g. using CGContextDrawImage(gtx, rect, img.CGImage)) and then extracting/copying the rendered component values from the rendered result over to a destination RGB bitmap.
If your inputs are not multicomponent color models (e.g. grayscale), then you should render to the source color model to save a bunch of CPU time and memory.