OS X, Cocoa: How to highlight burned-out areas of a photo? - image

Question

I've got an app that displays photos using NSImage – specifically, -[NSImage drawInRect:fromRect:operation:fraction:]. I want to highlight areas of the photo that are completely burned out (maximum values in all components, pure white) using a color like red, as some digital cameras and image processing apps do, to help the user see whether the image is overexposed, and how badly.
I've been scratching my head as to how to do this. Options I've considered:
I could probably write a Core Image filter to do it; none of the built-in filters look up to the task. That seems like overkill, though; I've been reading through the docs, and it looks fairly complicated. Big learning curve.
I could scan through the bitmap data for the image and modify it as necessary. This is easy enough to code for one bitmap format, but the multitude of bitmap formats make it a rather annoying exercise, and speed is important here, so writing general-purpose code that renders the image up to some maximal common format and works on that bitmap would be too big a speed penalty.
As it happens, I am already scanning through images (handling all the different bitmap formats) at an earlier point in the code, to generate histogram data for the images. I could pretty easily add code at that point that would remember the burned-out pixels for later use. I'm not quite sure what the best way is to do that, though. A 1-bit-per-pixel NSBitmapImageRep? How would I draw it later, making the 1-pixels draw red and the 0-pixels draw transparent, for example? I don't want to make a 32-bit NSBitmapImageRep with an alpha channel and everything just for this purpose, as memory is not infinite and images are large. But there must be a way to draw a 1-bit mask in a given color, somehow.
Before forging ahead with one of these approaches, I thought I'd see whether anybody here has a better idea. Or maybe has implemented the CI filter in question already? Apart from the learning curve, that seems like the best approach I've thought of so far – no memory overhead, and probably faster than other options, too.
Thanks...
Ben Haller
Stick Software
OK, I implemented my own Core Image filter to do this. Wasn't as hard as I expected, although the documentation is not great for this stuff. The doc examples all assume you're using ARC, so if you're not, following those examples will give you various retain/release bugs. There was also a little weirdness with the CIFilterConstructor stuff, which did not quite go as documented. But overall pretty easy. CI is cool. My code is below, for anybody who might find it useful:
Header:
#import
#interface SSTintHighlightsFilter : CIFilter
{
CIImage *inputImage;
CIColor *highlightColor;
}
#end
Implementation file:
#import "SSTintHighlightsFilter.h"
static CIKernel *tintHighlightsFilter = nil;
#implementation SSTintHighlightsFilter
+ (void)initialize
{
[CIFilter registerFilterName:#"SSTintHighlightsFilter" constructor:(id )self
classAttributes:[NSDictionary dictionaryWithObjectsAndKeys:#"Tint Highlights", kCIAttributeFilterDisplayName, [NSArray arrayWithObjects:kCICategoryColorAdjustment, kCICategoryStillImage, nil], kCIAttributeFilterCategories, nil]];
}
+ (CIFilter *)filterWithName:(NSString *)name
{
CIFilter *filter = [[self alloc] init];
return [filter autorelease];
}
- (id)init
{
if (!tintHighlightsFilter)
{
NSBundle *bundle = [NSBundle bundleForClass:[self class]];
NSString *code = [NSString stringWithContentsOfFile:[bundle pathForResource:#"tintHighlightsAndShadows" ofType:#"cikernel"] encoding:NSASCIIStringEncoding error:NULL];
NSArray *kernels = [CIKernel kernelsWithString:code];
tintHighlightsFilter = [[kernels objectAtIndex:0] retain];
}
return [super init];
}
- (NSDictionary *)customAttributes
{
NSDictionary *attrs = #{
#"highlightColor" : #{ kCIAttributeClass : [CIColor class], kCIAttributeType : kCIAttributeTypeOpaqueColor }
};
return attrs;
}
- (CIImage *)outputImage
{
CISampler *src = [CISampler samplerWithImage:inputImage];
return [self apply:tintHighlightsFilter
arguments:[NSArray arrayWithObjects:src, highlightColor, nil]
options:[NSDictionary dictionaryWithObjectsAndKeys:[src definition], kCIApplyOptionDefinition, nil]];
}
#end
tintHighlights.cikernel:
kernel vec4 tintHighlights(sampler inputImage, __color highlightColor)
{
vec4 originalColor, tintedColor;
float sum;
// fetch the source pixel
originalColor = sample(inputImage, samplerCoord(inputImage));
// calculate the color component sum as a way of testing whether we are black or white
sum = originalColor.r + originalColor.g + originalColor.b;
// replace pixels that are white with the highlight color
tintedColor = (sum > 2.99999999999999999999999) ? highlightColor : originalColor;
// preserve alpha
tintedColor.a = originalColor.a;
return tintedColor;
}
using the filter:
+ (NSImage *)showHighlightsInImage:(NSImage *)img dstRect:(NSRect)dstRect
{
NSGraphicsContext *currentContext = [NSGraphicsContext currentContext];
NSRect dstRectForCGImage = dstRect; // because the method below wants a pointer, and I don't trust it not to modify my rect...
CGImageRef cgImage = [img CGImageForProposedRect:&dstRectForCGImage context:currentContext hints:nil];
CIImage *inputImage = [[CIImage alloc] initWithCGImage:cgImage];
[SSTintHighlightsFilter class]; // get my filter initialized
CIFilter *highlightFilter = [CIFilter filterWithName:#"SSTintHighlightsFilter"];
[highlightFilter setValue:inputImage forKey:#"inputImage"];
[highlightFilter setValue:[CIColor colorWithRed:1.0 green:0.0 blue:0.0] forKey: #"highlightColor"];
[inputImage release];
CIImage *outputImage = [highlightFilter valueForKey:#"outputImage"];
NSImage *resultImage = [[NSImage alloc] initWithSize:[img size]];
[resultImage addRepresentation:[NSCIImageRep imageRepWithCIImage:outputImage]];
return [resultImage autorelease];
}
I'm not sure that I'm handling the alpha entirely robustly, with premultiplication issues and so forth, but apart from that possible glitch it is working great.