Count image entropy in swift using vImage on iOS/macOS

Question

I am trying to efficiently count Shannon entropy (the -sum(pi * log pi) formula) of an image on macOS/iOS with swift. I have found Accelerate framework and vImage functions, which look like what I am looking for, however documentation is scarce and I got lost in it.

I am creating vImage buffer like this

   var format = vImage_CGImageFormat(
        bitsPerComponent: 8,
        bitsPerPixel: 8 * 4,
        colorSpace: CGColorSpace(name: CGColorSpace.displayP3)!,
        bitmapInfo: .init(rawValue: CGImageAlphaInfo.noneSkipFirst.rawValue))!

    let buf = try vImage.PixelBuffer(
        cgImage: cg,
        cgImageFormat: &format,
        pixelFormat: vImage.Interleaved8x4.self)

My idea was to then convert it to 1 channel grayscale buffer ( vImage.PixelBuffer<vImage.Planar8> ) by buf.multiply(), according to this page: https://developer.apple.com/documentation/accelerate/converting_color_images_to_grayscale . Then create histogram from it, and then manually iterate over its 256 values and count the sum. However, it seems that vImage.PixelBuffer<vImage.Planar8> does not have histogram() method at all ... while vImage.PixelBuffer<vImage.Interleaved8x4> does.

Can you guide me to correct way to do it?

Answer 1

From my experience, I feel like your thinking is correct but we can streamline this a little more. These are some of the common steps that I have followed in previous occasions and got some insights from GPT as well.

Steps

1. Convert the image to grayscale

You can use the convert(to: ) method to transform a vImage.PixelBuffer from an interleaved colour format to a single channel.

2. Compute the Histogram

Use vImageHistogramCalculation_Planr8 to compute histogram info.

3. Calculate entropy

−∑(p i ⋅log(p i ​ ))

Example

Notes

• Make sure the image you are inserting is in the proper colour space.

This is somewhat of a rudimentary approach, there are a few other things you can add and make it more custom. But this solution will provide the necessary steps and output for the use cases in the selected custom info.

Feel free to reach out if you have any questions.