Many marked boxes are not correctly identified

Added by Mateo Obregon about 11 years ago

Hello-

I am testing the AMC system in hopes to use it for my teaching. In my tests, however, I have a lot of marked boxes that are not correctly identified as marked. I have usaed the suggested red boxes, together with the "remove red from scan". I have a recent multifunction printer/scanner and I can scan each page to a seperate TIF at 300dpi. I found that this was better than scanning multiple pages to a PDF.

My questions:

1. How should students mark the boxes so as to get the most reliable behaviour from AMC: Ticks, crosses, or filling in the boxes?

2. What values should I use for the "Scan" and "Project" parameters? In particular, "Black&White conversion threshold" and "Darkness threshold?

Mateo.


Replies (4)

RE: Many marked boxes are not correctly identified - Added by Alexis Bienvenüe about 11 years ago

I have a lot of marked boxes that are not correctly identified as marked.

Can you click on Zooms in the Data capture tab after having selected a page where a lot of boxes are badly detected, and post the window with the zooms?

RE: Many marked boxes are not correctly identified - Added by Mateo Obregon about 11 years ago

Thanks for your quick response. I have attached an example of what you ask for.

As you can see, some boxes have scores 0.38, 0.47 and (a dark) 0.87, but other boxes have scores of 0.009. I put the B/W threshold at 0.6 and the darkness threshold at 0.01 (as I cannot make it any smaller).

Mateo.

RE: Many marked boxes are not correctly identified - Added by Alexis Bienvenüe almost 12 years ago

AMC processes the scans as follows:
  1. first a grayscale to B/W conversion takes place (using the B/W threshold to separate white from black)
  2. then AMC counts the number of black pixels over the total number of pixels in each box. If this ratio is over the darkness threshold, then the box is declared to be ticked.

On the zooms window, the number written along with the boxes is the darkness ratio computed at step 2. Looking at your zooms, we can see that the darkness ratio is very low even if the box is fully darkened. So the problem comes from the first step: grayscale to B/W conversion. So you have to play with the B/W threshold value first, so that the ratios in the zooms window represents accurately the proportion of the box that is darkened. Please try a higher value for B/W threshold.

RE: Many marked boxes are not correctly identified - Added by Philipp von Klitzing almost 11 years ago

Best approach: Scan right away in b/w, not in greyscale.

(1-4/4)