Image Moment
A family of weighted sums over an image or contour used to summarize geometry such as area, centroid, orientation, and shape.
An image moment is a family of measurements rather than one feature. Low-order moments provide quantities such as area and centroid; central and normalized moments are used to separate translation and scale effects.
Hu moments derive rotation-invariant combinations from normalized central moments. "Invariant" does not mean noise-proof: small contour errors can noticeably perturb higher-order moments.
Practical Note
Shape descriptors change depending on whether moments are computed over grayscale intensities, a binary mask, or an extracted contour. That preprocessing path should be treated as part of the feature definition.
Source
- https://ieeexplore.ieee.org/document/1057692
- https://docs.opencv.org/4.13.0/d0/d49/tutorial_moments.html