Extract RGBA pixel data from an NSImage in Swift

NSImage+pixelData.swift

//
//  NSImage+pixelData.swift
//
//  Created by Figgleforth on 5/21/15.
//  Copyright (c) 2015 Bojan Percevic. All rights reserved.
//

import AppKit

extension NSImage {
   
   func pixelData() -> [Pixel] {
      var bmp = self.representations[0] as! NSBitmapImageRep
      var data: UnsafeMutablePointer<UInt8> = bmp.bitmapData
      var r, g, b, a: UInt8
      var pixels: [Pixel] = []
      
      for row in 0..<bmp.pixelsHigh {
         for col in 0..<bmp.pixelsWide {
            r = data.memory
            data = data.advancedBy(1)
            g = data.memory
            data = data.advancedBy(1)
            b = data.memory
            data = data.advancedBy(1)
            a = data.memory
            data = data.advancedBy(1)
            pixels.append(Pixel(r: r, g: g, b: b, a: a))
         }
      }
      
      return pixels
   }
   
}

struct Pixel {
   
   var r: Float
   var g: Float
   var b: Float
   var a: Float
   var row: Int
   var col: Int
   
   init(r: UInt8, g: UInt8, b: UInt8, a: UInt8, row: Int, col: Int) {
      self.r = Float(r)
      self.g = Float(g)
      self.b = Float(b)
      self.a = Float(a)
      self.row = row
      self.col = col
   }
   
   var color: NSColor {
      return NSColor(red: CGFloat(r/255.0), green: CGFloat(g/255.0), blue: CGFloat(b/255.0), alpha: CGFloat(a/255.0))
   }
   
   var description: String {
      return "RGBA(\(r), \(g), \(b), \(a))"
   }
   
}

Swift has changed a bit. This works for me as of June 2018.

extension NSImage {

    func pixelData() -> [Pixel] {
        let bmp = self.representations[0] as! NSBitmapImageRep
        var data: UnsafeMutablePointer<UInt8> = bmp.bitmapData!
        var r, g, b, a: UInt8
        var pixels: [Pixel] = []

        for row in 0..<bmp.pixelsHigh {
            for col in 0..<bmp.pixelsWide {
                r = data.pointee
                data = data.advanced(by: 1)
                g = data.pointee
                data = data.advanced(by: 1)
                b = data.pointee
                data = data.advanced(by: 1)
                a = data.pointee
                pixels.append(Pixel(r: r, g: g, b: b, a: a, row: row, col: col))
            }
        }

        return pixels
    }

}

struct Pixel {

    var r: Float
    var g: Float
    var b: Float
    var a: Float
    var row: Int
    var col: Int

    init(r: UInt8, g: UInt8, b: UInt8, a: UInt8, row: Int, col: Int) {
        self.r = Float(r)
        self.g = Float(g)
        self.b = Float(b)
        self.a = Float(a)
        self.row = row
        self.col = col
    }

    var color: NSColor {
        return NSColor(red: CGFloat(r/255.0), green: CGFloat(g/255.0), blue: CGFloat(b/255.0), alpha: CGFloat(a/255.0))
    }

    var description: String {
        return "RGBA(\(r), \(g), \(b), \(a))"
    }

}

Aren't you missing an 'data.advanced(by:1)' after grabbing the alpha?

let bmp = self.representations[0] as! NSBitmapImageRep // 3rd line
throws me an error so I changed for

let c = self.cgImage(forProposedRect: nil, context: nil, hints: nil)
        let bmp = NSBitmapImageRep(cgImage: c!)

and @hollandck was right

Aren't you missing an 'data.advanced(by:1)' after grabbing the alpha?

anyway, the most of @bpercevic 's code works pretty good, thanks a lot!

Fixed code (MacOS Catalina 10.15.5, Xcode 11.5):

import AppKit

extension NSImage {

    func pixelData() -> [Pixel] {
        
        let c = self.cgImage(forProposedRect: nil, context: nil, hints: nil)
        let bmp = NSBitmapImageRep(cgImage: c!)
        
        
//        let bmp =  self.representations as NSBitmapImageRep
        var data: UnsafeMutablePointer<UInt8> = bmp.bitmapData!
        var r, g, b, a: UInt8
        var pixels: [Pixel] = []

        for row in 0..<bmp.pixelsHigh {
            for col in 0..<bmp.pixelsWide {
                r = data.pointee
                data = data.advanced(by: 1)
                g = data.pointee
                data = data.advanced(by: 1)
                b = data.pointee
                data = data.advanced(by: 1)
                a = data.pointee
                
                data = data.advanced(by: 1)
                
                pixels.append(Pixel(r: r, g: g, b: b, a: a, row: row, col: col))
            }
        }

        return pixels
    }

}

struct Pixel {

    var r: Float
    var g: Float
    var b: Float
    var a: Float
    var row: Int
    var col: Int

    init(r: UInt8, g: UInt8, b: UInt8, a: UInt8, row: Int, col: Int) {
        self.r = Float(r)
        self.g = Float(g)
        self.b = Float(b)
        self.a = Float(a)
        self.row = row
        self.col = col
    }

    var color: NSColor {
        return NSColor(red: CGFloat(r/255.0), green: CGFloat(g/255.0), blue: CGFloat(b/255.0), alpha: CGFloat(a/255.0))
    }

    var description: String {
        return "RGBA(\(r), \(g), \(b), \(a))"
    }

}

I came across your gist while trying to track down some info that seems to be missing from Apple's docs. While it's often true that NSBitmapImageRep.bitmapData is a sequence of 8-bit RGBA values, it can also be the bytes of a sequence of big or little endian UInt16 or UInt32, or "floating point" - tracking down whether that always means Float or could possibly be Double is what I was actually looking for. The alpha channel can be first (and I'm not sure if that means ARGB or ABGR). What's more bitmapData could represent CMYKA (or possibly with alpha first). What fun, right?

My approach is to create a CGContextRef (drawing context) backed by my own data buffer and then draw the CGImage into that. That takes care of any different formats that @chipjarred mentions and may be faster as well, IDK. My Objective C++ code for that, which should be fairly easy to convert into Swift:

CGImageRef image = [ns_image CGImageForProposedRect:NULL context:NULL hints:NULL];  // macOS
  // or
CGImageRef image = ui_image.CGImage;  // iOS

if (image)
{
  int width  = (int) CGImageGetWidth(image);
  int height = (int) CGImageGetHeight(image);
  unsigned char* buffer = new char[ width * height * 4 ];
  CGContextRef gc = CGBitmapContextCreate(
                      (void*)buffer, width, height, 8, width*4,
                      CGColorSpaceCreateDeviceRGB(), kCGImageAlphaPremultipliedLast
                    );
  CGContextDrawImage( gc, CGRectMake(0, 0, (CGFloat)width, (CGFloat)height), image );
  CGContextRelease( gc );

  // 'buffer' now contains pixels as series of R,G,B,A bytes, which will be
  // ABGR when read as an Int32. I always swap B and R afterwards for my purposes.
}

@AbePralle: Nice way to do it. Basically you just let Core Graphics do the work of converting the image data to a known format for you. 'CGContextDrawImage' is fast, not to mention thoroughly debugged from decades of extensive usage. As for speed, if profiling were to indicate it's a problem for a given app, the main issue would likely be the dynamic memory allocation for the buffer; however, given that the gist's code repeatedly appends to an initially empty array, which has to re-allocate a few times before finally being big enough for the image data, I think your code would out-perform it. The gist's code could just call reserveCapacity before the loop to remedy that, and if it turned out to be an issue, both could use some kind of buffer/array re-use approach to avoid memory allocation overhead.