I have a rather large, almost full screen image that I'm going to be displaying on an iPad. The image is about 80% transparent. I need to, on the client, determine the bounding box of the opaque pixels, and then crop to that bounding box.
Scanning other questions here on StackOverflow and reading some of the CoreGraphics docs, I think I could accomplish this by:
CGBitmapContextCreate(...) // Use this to render the image to a byte array
..
- iterate through this byte array to find the bounding box
..
CGImageCreateWithImageInRect(image, boundingRect);
That just seems very inefficient and clunky. Is there something clever I can do with CGImage masks or something which makes use of the device's graphics acceleration to do this?
There is no clever cheat to get around having the device do the work, but there are some ways to accelerate the task, or minimize the impact on the user interface.
First, consider the need to accelerate this task. A simple iteration through this byte array may go fast enough. There may be no need to invest in optimizing this task if the app is just calculating this once per run or in reaction to a user's choice that takes at least a few seconds between choices.
If the bounding box is not needed for some time after the image becomes available, this iteration may be launched in a separate thread. That way the calculation doesn't block the main interface thread. Grand Central Dispatch may make using a separate thread for this task easier.
If the task must be accelerated, maybe this is real time processing of video images, then parallel processing of the data may help. The Accelerate framework may help in setting up SIMD calculations on the data. Or, to really get performance with this iteration, ARM assembly language code using the NEON SIMD operations could get great results with significant development effort.
The last choice is to investigate a better algorithm. There's a huge body of work on detecting features in images. An edge detection algorithm may be faster than a simple iteration through the byte array. Maybe Apple will add edge detection capabilities to Core Graphics in the future which can be applied to this case. An Apple implemented image processing capability may not be an exact match for this case, but Apple's implementation should be optimized to use the SIMD or GPU capabilities of the iPad, resulting in better overall performance.