Bag of feature : I want to retrieve target-image from training images

I want to retrieve target-image from training images, first, I detail the descriptor but finding nothing useful. The descriptor's structure is like:
Mat {
step: 512,
elemSize: 4,
sizes: [ 2000, 128 ],
empty: false,
depth: 5,
dims: 2,
channels: 1,
type: 5,
cols: 128,
rows: 2000 }
so how can I get feartures' Matrix form an image ? Please Help! THX!

The issue mentions only an OpenCV Mat descriptor and no files, tests, or entry points. First clarify which training and target-image retrieval behavior is expected, then inspect the repository's OpenCV and Node.js binding surfaces to identify where such a feature would belong and how completion could be verified.

Written by the indexing model from the issue text.