Image Pyramids refers to the downsampling of images at different levels to obtain images with different resolutions.
In the Feature Matching algorithm, we can match images on different layers to achieve scale invariance. For example, if the camera moves backward, we should find a match in the upper layer of the previous image pyramid and the lower layer of the next image.
From Cyrill Stachniss
An image pyramid is a list of images. Each image has half of the size of its predecessor