Poor sensitivity on scans: due to ovals?

Added by Devin Casenhiser over 8 years ago

I've noticed that when I've given the last two exams, the sensitivity for nearly all the scans are at or near 10. MSE is fine, however. I'm not sure what is different. I'm running 1.2.1495. I also think that I switched at the same time to ovals rather than squares. Could this be the problem?

Or perhaps a scan setting needs to be adjusted?

Scan settings are as follows:

Vector formats dpi: 300
B&w conversion threshold: .50

Marks size max increase: .20
Marks size max decrease: .20
default darkness threshold: .20
measured box proportion: .80

process scans with 3 corner marks: checked

Cheers!
Devin


Replies (4)

RE: Poor sensitivity on scans: due to ovals? - Added by Devin Casenhiser over 8 years ago

I notice in the zooms that the boxes that aren't correctly identified as being filled in have the value .000 next to them (for the most part). They are definitely filled in, however. No question about it.

RE: Poor sensitivity on scans: due to ovals? - Added by Alexis Bienvenüe over 8 years ago

Can you post a screenshot of a zooms window for a problematic scan?

RE: Poor sensitivity on scans: due to ovals? - Added by Devin Casenhiser over 8 years ago

Hi Alexis,

I seem to have been able to correct the problem by deleting and then re-importing the scans (pressing the automatic button). Not sure why that made a difference, but it did. I still get the sensitity column mostly in red, but for the most part the scans are correct.

Screen shot attached.

RE: Poor sensitivity on scans: due to ovals? - Added by Nikola Z. Guscic about 8 years ago

Were your exams printed or photocopied?

In my experience, small ovals work fine when printed, but can be problematic when photocopied because the process of photocopying
introduces slight distortions in the page size and proportions. You can often see when you superimpose the original and the photocopy that one might have slightly longer diagonal between page markers.

Normally it is not a problem, especially if you tweak the 'measured box proportion setting' to be a bit lower than default, say 0.65. However, with very small ovals, those small distortions are significant enough to cause some checked ovals to be unrecognized (about 2-3% in my experiments).

It is easy to get around this by simply increasing the size of ovals (this doesn't change the layout of the answer sheet).

(1-4/4)