Training a cascaded classifier while including image annotations also as features

Question

I understand that OpenCV's cascaded classifier trains on images and uses the features that the cascaded trainer itself extracts from the image. But is it possible to add into the training some annotations on the images? For example, I have around 600 images to use as positive samples to train the classifier, but I need to annotate each image with a short vector of additional features, such as ("aerial view", "planar view", or "city background" or "landscape background").

Is it possible to build a classifier that combines the trainer's extracted features as well as manually supplied annotations?

Answer 1

This will not work in this case, because the same features used for training must be available for classification. Let's say you train a car classifier using annotations like "side view" and "front view" as features. Then, when you try to classify an unknown image, you would need to supply the annotation too, which mostly defeats the purpose of your classifier.

The point is that you need to be able to extract the features automatically. If the input to the classifier is an image, then the features must be computable from the image.

Having said that, what you are suggesting can work if the input to your classifier is an image and some associated text. Maybe a caption from a picture in a newspaper or a figure in a text book. Then you can extract some features (e. g. keywords) from the text as well.