[Standard]: Biased estimator of Standard Deviation

[LarsKooijman]

Member Body/National Committee

AS ISO IEC 29794.4:2025

Clause/Subclause

6.2.16.3

Paragraph/Figure/Table

No response

Type of Comment

Technical

Comments

The section describes an unbiased estimator of the standard deviation. Specifically, N x M is corrected by a minus 1.
However, C++ code utilises OPENCV_CORE_HPP function meanStdDev, which seems to employ a biased estimator -> link.

Proposed Change

I would opt for a code change to ensure the estimator is unbiased for standard deviation as is described in the norm.

[gfiumara]

Discuss with group, but the $N \times{} M\mathbf{-1}$ does seem to imply unbiased. If we proceed, we need to make this more explicit using words in the text, and can likely use Bassel's correction (stddev * std::sqrt(N / (N - 1.0))). We might want to see how much this actually changes things in the compliance test and avoid the extra math if unnecessary.