detect surface defects

[durnek60]

Hi all!

First of all, Merry Christmas! by the way, the work never ends. I am wondering about a solution, without any success. Could you help me?

My problem is the following. I have a surface, and I have to detect all of defects on it. I don't have any good idea how to find them. In the attachment, you can see a small surface defect. My project is: find and localize all of these defects automatically, and measure them. What kind of image processing tools can be used to find and measure (area) them?

Measuring them is not as important as than finding, and localizing them.

First of all, using a morphological procedures could be a good start, but I don't know to detect these defect automatically...

Please see the attachment:

[crelf]

Wow - that's a tricky one due to the very very low contrast. My first instinct is to suggest a different setup - if the item that you're interested in is the tan/pink thing that takes up te lower part of the image and the gray/white part at the top is just a background, then I'd change your lighting - you're using lighting that's on the camera side of the object you're looking at, but I'd go for a back light - remove all the lighting (or, at least, limit it) from the camera side, and out the item on a backlight (something like one of these from NERlite) - that will give you a high-contrast image that you can then binarize with an appropriate threshold - that will make the defect really stand out - then you can maybe determine the angle of the edge, and then try a parallel rake that's perpendicular to the edge.

Shameless plug: You can find more infomation about all of these techniques, and more, in Image Acquisition and Processing with LabVIEW.

<rant>

You could try to solve the issue in software, but, if you can do it in hardware, you definately should! I've seen far too many setups where engineers have tried to apply countless software image transforms on top of other transforms to, essentially, get an image that the right camera/lense/lighting combination would have gotten much more consistantly. If you had a thermocouple on the end of 200 meters of noisy cable, you wouldn't try to filter out the noise in software - you'd fix the hardware setup. Image acquisition and processing is the same. As my old DAQ professor used to say: garbage in, garbage out

</rant>

[durnek60]

 Thank you very much! 

In the middle of January I will start to develop any solution for this problem. Yes, you are totally right about lightening. Changing it  will be the first thing what's need to be replaced... 

If I have any problems, I gonna tell you. BTW, i will share my solution too 

Best Regards!

Edited December 27, 2009 by durnek60

[crelf]

In the middle of January I will start to develop any solution for this problem.

Great - if you'd like, you're welcome to upload or describe your design as you're designing it and we'll critique it for you as you go.

[durnek60]

Great - if you'd like, you're welcome to upload or describe your design as you're designing it and we'll critique it for you as you go.

OK! Thanks for the possibility!!!