When you run some process, there are two ways to look for anomalies – “white box” and “black box”. First one uses some knowledge about the process. Perhaps, you are already doing that. But comprehensive white box analysis is very expensive, so it is easy to overlook something. How about doing black box analysis on top of it?

I wrote some software that does black box search for anomalies in a series of images. It does Fourier transform, crops the low frequency part, calculates average, and then calculates distance from the average for each image.

Here is the focus sweep for the bug:

Here is a plot for some cyclical process I study:

This software also generates sorted html file with links to images, so you can quickly look at anomalies:

Here are C# sources
You are welcome to download and build it. If you just want to use it, here are Windows (.NET) binaries (just unzip files to some folder).

Here is how I run it:
“C:\Anomalies\Anomalies.exe” “C:\PHOTO\Bug” “jpg”
where C:\PHOTO\Bug is a folder with jpegs.
Images could be color or grayscale. PNG and other image formats are supported too (whatever OpenCV can load).
It also supports recursive folders.