Kermit developed this image processing app using OpenCV Python and scikit-learn. Users input an image or take a picture with their camera. The app then gives a prediction of which Kanto Starter Pokemon (Pikachu, Bulbasaur, Charmander, and Squirtle) it believes the image is. If the app can't determine, it outputs the Pokemon "Ditto" instead.
Kermit created his own dataset of training data, testing data, and validation data. He then trained an SVM (support vector machine) for each Kanto Starter Pokemon, as the main machine learning model. The app uses HOG (histrogram of oriented gradients) as the main feature extractor to decide major feature points of each Pokemon, for instance, the bulb on Bulbasaur, or Squirtle's shell. That way, the predictions aren't too affected by RGB color or image orientation.
