Waldoko is funny side project I imagined during my 2D signal processing class. I wanted to crack the famous game "Where's Waldo" using image processing techniques. The idea is to use a pattern matching algorithm to find Waldo in the image.
Why "Waldoko"? Because it's a mix between "Waldo" and "doko" (Japanese for "where").
The project is divided into two main parts:
- Pattern Building: We need to build a pattern from the image of Waldo. This pattern will be used to find Waldo in every other images. So from the very start, from learning how to open a picture in Python, to the final pattern, we will go through the following steps: - Open the image - Convert it to grayscale - Detect the edges - Find the contours - Draw the bounding box - Extract the pattern
It was the plan, but I'm lazy and I've done it manually. Check the 1st ipynb for more details. I'm letting room for improvement here, but I'm not sure I'll ever come back to this project.
- Pattern Matching: We will use the pattern to find Waldo in the other images. I used the phase correlation algorithm to find the pattern in the images. The phase correlation is a technique used to find the translation between two images. It's often used in image registration, but it can also be used to find a pattern in an image.
I've tested the algorithm on a few images and it worked pretty well but to be honest, a good old CNN would have done the job better. But it was fun to try something different.
