If you want to memory map the encoded image data, then mmap a file and provide a reference to the data by passing a CGDataProviderRef to CGImageCreate.
mapped = mmap( NULL , length , ... );
provider = CGDataProviderCreateWithData( mapped , mapped , length , munmap_wrapper );
image = CGImageCreate( ... , provider , ... );
uiimage = [UIImage imageWithCGImage:image];
...
Where munmap_wrapper is something like this:
// conform to CGDataProviderReleaseDataCallback
void munmap_wrapper( void *p , const void *cp , size_t l ) { munmap( p , l ); }
If you want to memory map the actual pixels, instead of the encoded source data, you would do something similar with a CGBitmapContext. You would also create the provider and image so the image refers to the same pixels as the context. Whatever is drawn in the context will be the content of the image. The width, height, color space and other parameters should be identical for the context and image.
context = CGBitmapContextCreate( mapped , ... );
In this case, length will be at least bytes_per_row*height bytes so the file must be at least that large.
If you have an existing image and you want to mmap the pixels, then create the bitmap context with the size and color space of your image and use CGContextDrawImage to draw the image in the context.
You did not say the source of your image, but if you are creating it at runtime it would be more efficient to create it directly in the bitmap context. Any image creation requires a bitmap context behind the scenes, so it might as well be the memory mapped one from the start.