An OpenCV based system that uses a Haar Cascade Classifier to detect facial features of Japanese anime characters in a given image file.
-
Linux Distro
-
Python 3.4+
-
OpenCV 3.2.0+ (If using a binary, ensure GTK support is enabled.)
-
googleimagesdownload (Install using
pip3 install googleimagesdownload
) -
saucenao (Install from here)
-
Get positive images for the characters you want to train.
-
Get negative images for the classifier.
-
Create more positive samples by adding convolutions to the positive images and superposing them on the negative images.
-
Train the classifier using the in-built OpenCV commands.
-
Add images you want to test in the
test_set
folder. -
Run
python3 detect.py test_set/<image-name>
. The script should detect the image and scrape similar images using Google Images and saucenao.
This article was very informative in explaining the training process for the Haar Cascade classifier.