Category Archives: spectral response

Minimalist ESF, LSF, MTF by Monotonic Regression

Because the Slanted Edge Method of estimating the Spectral Frequency Response of a camera and lens is one of the more popular articles on this site, I have fielded variations on the following question many times over the past ten years:

How do you go from the intensity cloud  produced by the projection of a slanted edge captured in a raw file to a good estimate of the relevant Line Spread Function?

Figure 1.  Slanted edge captured in the raw data and projected to the edge normal.  The data noisy because of shot noise and PRNU.  How to estimate the underlying edge profile (orange line, the Edge Spread Function)?

So I decided to write down the answer that I have settled on.  It relies on monotone spline regression to obtain an Edge Spread Function (ESF) and then reuses the parameters of the regression to infer the relative regularized Line Spread Function (LSF) analytically in one go.

This front-loads all uncertainty to just the estimation of the ESF since the other steps on the way to the SFR become purely mechanical.  In addition the monotonicity constraint puts some guardrails around the curve, keeping it on the straight and narrow without further effort.

This minimalist, what-you-see-is-what-you-get approach gets around the usual need for signal conditioning such as binning, finite difference calculations and other filtering, with their drawbacks and compensations.  It has the potential to be further refined so consider it a hot-rod DIY kit.  Even so it is an intuitively direct implementation of the method and it provides strong validation for Frans van den Bergh’s open source MTF Mapper, the undisputed champ in this space,[1] as it produces very similar results with raw slanted edge captures. Continue reading Minimalist ESF, LSF, MTF by Monotonic Regression

COMBINING BAYER CFA MTF Curves – II

In this and the previous article I discuss how Modulation Transfer Functions (MTF) obtained from every color channel of a Bayer CFA raw capture in isolation can be combined to provide a meaningful composite MTF curve for the imaging system as a whole.

There are two ways that this can be accomplished: an input-referred approach (L) that reflects the performance of the hardware only; and an output-referred one (Y) that also takes into consideration how the image will be displayed.  Both are valid and differences are typically minor, though the weights of the latter are scene, camera/lens, illuminant dependent – while the former are not.  Therefore my recommendation in this context is to stick with input-referred weights when comparing cameras and lenses.1 Continue reading COMBINING BAYER CFA MTF Curves – II