Image Dataset Creator and Normalizer for Machine Learning Projects
This is a series of tools to create image datasets for machine learning based on google image search and a pre-trained neural network to discriminate useful or rubbish images.
The tools I created are:
- GoogleImageSpider.py: A wrapper-spider which function is to emulate a browser using google image search in order to download a bunch o images about the same topic
- ImageNormalizer.py: Really useful class to normalize PNG, JPG or GIF of images into a little thumbnails of the size, background and no transparencies images finally saved in a JPG. (Really useful for my ML projects)
- FaceClassifier.py: Using OpenCV we can detect faces, recognize them and crop to a size to normalize from the pictures you got from GoogleImageSpider. You can see a demo with Trump here.
- ImageClassifier.py: Last but not least, an CNN implementing an image classifier in order to tag "GOOD" and "BAD" images with two or tree simples lines of code.
here are some examples about how all this works trying to create datasets
Pokemon dataset example and how the final result will be:
And this another one for trump faces at trump faces dataset:
