Matching analysis in case of LCD

[anusree]

Hello,

I am working on one of the project which needs to check for the segments displayed on the LCD and compare whether the captured image is same as the template. When I do the pattern mathing using IMAQ find pattern 2 in which captured image and template image as two image input the result obtained is not as required . The analysis is to check whether the LCD segments are lighted in the pattern as required. Even if one of the segments doesn't light up it has to give us the result as false. how can i do the matching in this.

With regards,

anusree

[Ken Pietrzak]

Hello anusree,

It would be helpful for us if you could post a few sample images of the LCD and the templates that you have created.  I assume that you are training your template from a "good" LCD.

Thanks,

Ken Pietrzak

www.displaycheck.com

[anusree]

Hello ken,

I have attached th etemplate image.

[anusree]

Hello ken,

This is the second image in which the dot in third segment is missing, we need to identify any small segment change in the display  and result as fail.

[Logic]

Hi anusree,

As far i know, simply pattern matching for this will lead to failure. It may give a different score for the dot is not present image, but that will also be not reliable over the time.

My idea is to threshold the image and store the particle position and size of the each particle in the good image, and check for the same in the every image. That might work well. Just a thought, see whether works.

thanks,

L0G1C

[chilly_charly]

A simple solution to detect missing segments is to do an Exclusive-OR logical operation between the two images. A threshold then isolate the faulty pixels. This works because the position of the display is exactly the same in both pictures, and because the lighting conditions are very similars.

The IMAQ find pattern is best suited when the object position is random, for instance pieces on a conveyor belt. But here, you can develop a better suited solution.

See the attached vi.

[Ken Pietrzak]

In a similar vein, you can also simply subtract the template image from the sample image if you are always looking for missing black segments.  This would be a bit cleaner than Ex Or. 

If you really think you need to use pattern matching, I would suggest breaking down your template image into a set of smaller templates.   For example, make a template of just the N and search for that N - making sure that you find one and only one N with a score above a score that would be achieved with an N with a missing segment.   Then create a template for each type or group of vertical segments and get a count for each of those that you find above a set threshold.  You may want to make a template of the dot alone or the dot with two neighboring vertical segments as another template to search for.

Ken Pietrzak

www.displaycheck.com

[anusree]

Hi,

As I worked by subtracting the images or by Xor two images error pops up  "Error -1074396074 occurred at IMAQ Xor
  Incompatible image size" because the comparision is between the template image and the captured image from the camera. The camera captures whole visible area but we need to analyse on the LCD part. pls suggest me in this.

[chilly_charly]

Remember that we can't help you if you don't provide enough information.

The image size of the pictures you posted before was exacly the same. What procedure did you apply to obtain this result ? Why can't you do the same now ?

I also noticed that these pictures were RVB images, while they are clearly grey level images. Could that be the source of your problem ?

[anusree]

Hello CC,

I am new in using Vision. I have captured image of the device under test from MAX using camera and saved as *.jpg and extracted the LCD part and made as a template file.

In my application I need to check weather the LCD of the device is Ok or Fail

If the segments lighted is similar to the Template the LCD is ok else fail even a small change like half segment lighted, dot missing , letter missing in the display should be able to detect .

The captured image is attached

anusree