I’m currently trying to create a real-time CIImage-based filter for a captured video-stream. The basic task is this: take a CIImage of size (sx, sy) and apply a mapping that yields an CIImage of size (dx, dy), where dx > sx, dy > sy.

The mapping is basically an array of coordinates with dimension (dx, dy). At each point, the source-pixel is stored. This is the inner loop of my mapping, just to make you grasp the idea:

for(int dy = 0; dy < _dest.height; dy++) { coord *mappingRow = _mappingRows[dy]; unsigned int *pixelRow = _dest.rowtab[dy]; for(int dx = 0; dx < _dest.width; dx++) { int sx = mappingRow[dx].x; int sy = mappingRow[dx].y; *(pixelRow++) = _source.rowtab[sy][sx]; } }

So far, so good. The _source and _dest-variables are structs that contain all information necessary to deal with CIContext and look like this:

typedef struct { CGContextRef port; CIContext* context; int width; int height; int bytesPerRow; int byteCount; CIFormat picelFormat; void * data; int **rowtab; } ContextInfo;

They are initialized in the init-method of my filter-object, as well as the mapping-table.

Everything works just fine when using the filter once. This is the whole map-method, containing the above loop. I draw the image to _source, map it to _dest.data and create an CIImage out of that.

• (CIImage ) map: (CIImage)inputImage { CGRect extent = [inputImage extent];

  [_source.context drawImage: inputImage atPoint: extent.origin fromRect: extent]; for(int dy = 0; dy < _dest.height; dy++) { coord *mappingRow = _mappingRows[dy]; unsigned int *pixelRow = _dest.rowtab[dy]; for(int dx = 0; dx < _dest.width; dx++) { int sx = mappingRow[dx].x; int sy = mappingRow[dx].y; *(pixelRow++) = _source.rowtab[sy][sx]; } }

  CIImage * r = [[[CIImage alloc] //initWithBitmapData:[NSData dataWithBytes: _dest.data length: _dest.byteCount] initWithBitmapData:[NSData dataWithBytesNoCopy: _dest.data length: _dest.byteCount freeWhenDone: NO] bytesPerRow: _dest.bytesPerRow size:CGSizeMake(_dest.width, _dest.height) format:kCIFormatARGB8 colorSpace:_colorSpace] autorelease]; return [ r autorelease ]; }

As you might see, the raw pixel buffers _source.data and _dest.data are allocated once and are supposed to be worked on as long as the filter exists.

Now when using this filter in my sequence-grabbing application that captures an NSImage from my i-Fire camera, I do this in an extra method:

-(NSImage) mapWithNSImage: (NSImage) image { NSBitmapImageRep *bitmap = image representations] objectAtIndex: 0]; [[CIImage *im = [[[CIImage alloc] initWithBitmapImageRep: bitmap] autorelease];

im = [self map: im];

bitmap = [NSCIImageRep imageRepWithCIImage: im]; CGRect extent = [im extent]; image = [[[NSImage alloc] initWithSize: NSMakeSize(extent.size.width, extent.size.height)] autorelease]; [image addRepresentation: bitmap]; return image ;

}

See also (http://www.gigliwood.com/weblog/Cocoa/Core_Image__Practic.html)