find missing dot

[stww]

Hello, I need to analyse picture from camera. Camera is looking at LED screen panel and trying to identify missing (not working) LEDs.

Now my program is able to count number of LEDs and make decision if screen is working. Next step for me is to identify which LED is not working. (position of missing white point at picture) there could be 0 and also 100 mising points 😞

I am sending my current program and picture from camera (after BCG correction).

I was trying to work with Particle reports output which contains positions of all found bright points, bud LED screen is not at 0° angle, so array dont have values is row. Do you have some idea how to do it? i also tryed detection of defects but it didnt worked.

[wiebe@CARYA]

I think there are dozens completely different solutions to this problem... These are just a few (partial) solutions that comes to mind...

If the distance between the dots is known, you can draw a slightly bigger rectangle on all the positions of the dots. Then threshold the image, so all black pixels are true. A particle analysis would reveal the locations of the broken pixels. Not very sophisticated, it's rather blunt, but could prove to be very robust. Complication for neighboring broken pixels...

All the positions are on a rotated grid. Statistics would give the rotation of the grid. Once known, you can position each particle on the grid, and the missing locations would be known afterward.

The broken pixel is a big black square. A template\pattern match would reveal the locations. Again complication for neighboring broken pixels...

A caliper could be used to calculate the rotation angle. Compensating the image, or compensating the found coordinates will easy the problem.

[stww]

Thanks for reply. I know that problem is really complex and i am also thinking about some kind of compensation of image or coordinates. I was looking for some solution but didnt found nothing. can you explain please? 

maybe corrected image with dots is grid will help

[Bob_Schor]

Have you looked in the LabVIEW Help, especially NI Vision Concepts Help?  Have you read about Spatial Calibration?  Are you familiar with mathematical transforms (to correct for perspective, rotation, etc.)?  If not, you should seek help from someone who likes math, vectors, matrices, etc.  If you have someone in your area with digital image processing experience, they should also be useful to consult.

Bob Schor

[billko]

As an aside, are you looking for hot pixels as well?

[stww]

for hot pixels? i tihnk i dont understand your question. please describe more

[billko]

Pixels stuck on.  It happens with LEDs, too.

[stww]

now i understand. Yes this is also checked, bud in different process, not necessary to do it here. 

I was also thinking about some kind of camera callibration to make image scanning correct. But I prefer some method to identify points and calculate near points, but I need some idea how to do it 

[wiebe@CARYA]

---

@stww wrote:

But I prefer some method to identify points and calculate near points, but I need some idea how to do it 

---

Not sure what you mean by that. The points are identified by their coordinates. Finding point near each other can be done by calculating the Euclidian distance.

That would give a rotation as well, although it would be inaccurate. You would want to calculate the rotation from points as far from each other as possible, so the measuring error gets less significant. You could try to get a rough rotation, rotate all the points, and then get two points far from each other on the same horizontal line (which is easier after the rough rotation) and calculate the accurate rotation from them...

You need to define your problem very well, and focus on that. With vision it's very easy to wander off to side tracks until you're lost... Not sure if that's what's happening, but getting the angle of rotation for instance might not be required at all.

Finding missing dots is not that hard. Simply dilate all "good" dots until they connect. The big black blobs are the missing dots, easy to find and label the objects, and thus the missing dots.