Negative images for haar cascade classifier training for facial expressions

Question

I'm trying to create my own cascade classifiers to detect facial expressions in images using OpenCV. I know that I need to collect both positive and negative images, where the positives contain the object to be detected, and the negatives do not. In face detection, the negatives are images where there are no faces. In order to create a cascade classifier for a certain facial expression, should the negatives be people with neutral expressions, or just images without faces in them?

The goal is to detect a few expressions (neutral, smiling/laughter, excitement, bored/sad).

Answer 1

"I'm trying to create my own cascade classifiers to detect facial expressions"

that probably won't work. you'd have to train one cascade per expression, and they are just not different enough for this.

instead, try to train a FisherFaceRecognizer on your expressions in a similar manner to this