Last Wednesday I had the opportunity to attend a workshop about CoreImage from Tobias Due Munk about Core Image.

First of all I would like to point how kind he was, as he sent me the material the night before as I told him I would arrive halfway through the workshop 👏🏽👏🏽👏🏽

It was really interesting, expecially as i never really went too much into Core Image apart using a standard CIFilter on a UIImage, so I wanted to wrap my head around 3 little things that I would like to takeaway from it.

Type-safe Core Image

While Core Image is a pretty old frameworks whose APIs were initially all stringly typed, Tobias showed us how to make use of newer, safer APIs.

Originally I would create and configure a CIFilter like this:

import CoreImage

let filter = CIFilter(name: "CIPhotoEffectNoir")
filter?.setValue(myImage, forKey: "inputImage")
let result = filter?.outputImage

Now, this code works, but it uses strings to both create and configure the filter. Additionally, it forces us to deal with an optional CIFilter? even though we know the filter is exists.

The trick is adding a new import statement:

import CoreImage
import CoreImage.CIFilterBuiltins

let filter = CIFilter.gaussianBlur()
filter.inputImage = myImage
let result = filter.outputImage

This results in great improvements, as we can now

  • rely on auto completion to find available filter names
  • set parameters as normal properties on the object
  • avoid dealing with optionals!

Image recognition

Core Image has an initialiser for a CIImage that takes a dictionary [CIImageOption: Any], setting true/false as the value for an option key enables/disables the option.

Here is an example of using an option to get the silhouette of a portrait:

let options: [CIImageOption: Any] = [
    .auxiliaryPortraitEffectsMatte: true
]

let ciimage = CIImage(image: myImage, options: options)

But there is many many options to get stuff like hairs, teeth, skin etc. Some more documented, some…😅

Metal

Last but not least, apparently you can use your own metal logic instead of using built in filters.

One example of a Metal kernel is CIBlendKernel whose Metal function is a function that gets called once per pixel and has the foreground and background images as inputs and the pixel to use in the result image as output.

Colors are of float4 type (or the type alias sample_t), which is a vector that has red, green, blue and alpha as values.

Really looking forward to explore more!