Listen to a special audio message from Bill Roper to the Hive Workshop community (Bill is a former Vice President of Blizzard Entertainment, Producer, Designer, Musician, Voice Actor) 🔗Click here to hear his message!
As a programmer with some image processing knowledge I would say the images could be produced by the following steps programmatically.
Gather all WoW icons in a way they can be easily found, eg the same folder.
For each WoW icon, find its average colour. This may need gamma correction per sample based on how they are stored.
For each WoW icon and its average colour you insert it in a data structure that allows searching (some sort of tree I would imagine, maybe one for each colour component). This forms an image dictionary.
For each pixel in your desired image query your image dictionary for the approximate matches (some amount of tolerance for variance). Add all results to a list.
Choose a random element from the list (maybe a least used bias to stop repetition) and use that icon to represent the source pixel in the product image. This is repeated until image is finished.
Improvements to the algorithm might include trying to match areas with icons most closely resembling the area. This is an expansion on the method outlined above where instead of using a single sample per icon and pixel in the resulting image, multiple samples are used. This solves variation problems and improves result quality as it uses more unique information to choose which icon is used where (allowing for effectively a higher resolution product image). This adds a lot of complexity though and might require a very complex data structure to be efficient.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.