The spectral signal-to-noise ratio resolution criterion: computational efficiency and statistical precision

This note describes a practical improvement in the computational efficiency of the spectral signal-to-noise ratio (SSNR) resolution criterion for correlation-averaged images. The total set of N images is randomly partitioned into ng subsets, each subset is separately averaged, and a reduced form of the SSNR is computed from these average images. In general, larger values of ng achieve lower statistical uncertainty, while smaller values of ng are computationally more expedient. It is shown that, for negatively stained data, a judicious compromise is achieved with 10 less than or equal to ng less than or equal to 20, regardless of how large N may be.