hi,

i still am starting with the project related to image processing. i want to make an automatic checker and yet i still am trying to look for a site that would be a great help with regards to the hardware.

i am still working on how would this work.

could you suggest any helpful sites that could really help me with the hardware portion of this project??? pls. let me know. i've been looking and yet found somethig quite not helpful.

thankz...

 

By "automatic checker," do you mean "bar-code scanner?"

 

"Image processing". Are you looking for suggestions about cameras?

 

And what is an OMR? You didn't mean OCR by chance?

 

optical mark recognition... i'm working on an automatic checker for multiple choice questions...

 

having the formed scanned or captured in a camera, the difficulty starts butting in... i am really having a hard time...

 

If the two images are perfectly aligned, then XORing every pixel between the two will indicate only the differences between them.

 

Alignment wouldn't have to be "perfect," just less than 1/2 pixel off. If the pixels were 1/2 or 1/3 as big as the answer marks, or even a bit larger than the answer marks, alignment would be fairly easy.

I've taken lots of exams with an alignment key down the edge of the page. I was told at one point the scanner used CSDs. There are no doubt other schemes used by other scanners.

 

Two channels of photoreceptors should give you what you need. One channel looks for an allignment strip that indicates proper feeding and also a marker for the answer boxes. The other looks at the answer boxes and indicates filled-in or blank. If that signal occurs when the allignment sensor is active, the coincidence means there is a good detection.

 

IIRC, there was a big deal made about #2 pencils - something about how the light reflects off the graphite. Harder or softer pencils would not always be read correctly.

 

What we are talking about IMO is pattern recognition. The way the robots do it at work is look for fudicials, special marks the machine looks for and aligns to. Then everything else falls into place.

 

I am interpreting this as scoring test sheets. The allignment channel catches the special marks. It might be worth investigating light transmission as opposed to light reflectance.

 

are there any helpful software that you could suggest on how to process or work on images aside from CVIP tools?

thankz!!!

 

any suggestion on how to start this project...

thank you...

 

It's not too realistic to just do this with image processing. Every such scanner I have seen/used moved the answer sheet through guides with powered rollers. The one detector looked for registration marks to verify proper orientation. The other looks for light (reflectance or transmission) being obscured by a mark in one of the answer boxes.

Programming tells the instrument how many answer fields are on the form, and how many answer boxes are in each field. Only one box may be filled in in each field.

The total project is partly mechanical to make the scan rollers and guide. It's partly involved with having answer forms made up to suit the scanner - more than one column of answers is quite possible. And then there is the means of instructing the microprocessor about the form in use so it can scan the answers.

 

There exist computer methods for pattern recognition even from tilted and completely unalgined photographs, but it is not really easy for implementation.
For aligned photos you could maybe use Bayes classification or somethnig else.

 

My suggestion would be to hack or reverse engineer something like a HP mfp (multifunction printer, "all-in-one" printer). The LJ4100 series has auto document feed, scans by moving the scan head, not the document, and image storage/transmission capability. It is also network enabled...you don't need a print server. It is a little tricky to download the images from, but once done, which means getting the right software, they can easily be processed.

The LJ4100 is several years old. There are certainly newer, cheaper
models available. When one considers the mechanics involved in an MFP, it is a little hard to justify trying to make one for the same price. You can then concentrate on the software.

John

 

Another option might be to make a hand-held scanner. The teacher (or teacher's assistant) could manually run the scanner down the appropriate side of each page.

Still another option might be to somehow adapt a bar-code scanner to read the dots.

 

hey, what has made the bar code scanner work???

thank you...

 

thanks for the Bayes classification tip...