VBAI

[Brad]

Thanks for the sample images, they were very helpful. One last question, do you want the ROI to change each time based on some edge locations of the current part, or do you want to have a single ideal ROI that uses a coordinate system to position over the current part? It seems like if you use edge detection or some other technique to create an ROI that changes for each part, you could miss defects along the edge since this may change the shape of the ROI to not include those defects. 

I was also curious in the first NG image if the faint 45 degree angle scratch is the cause for the part to not be good. Just wanted to make sure I knew why that one failed.

Thanks,

Brad

[Brad]

I also wanted to see if you could send a zip with individual image files instead of using screenshots of the images in a pdf. This will make it easier to provide a sample inspection that illustrates how you can process the images.

Thanks,

Brad

[minhtuanbk49]

Dear Brad,

Thank you for your share. I want a ROI change each time based on edge of sample (Of course ideally it should be the straight line, if any defect appears on this line I hope we can detect later.

For abnormal defect, I normally define any area with different color and bigger than 20 pixel (people can see) as defect.

The image is quite big, I wish I can send you by mail. Can you share for me your gmail?

[Brad]

You can post images here (feel free to zip and password protect if you want).

Copy this path indo a Windows file explorer (i.e. not internet explorer but file browser) and then you can just drag/drop the file.

ftp://ftp.ni.com/incoming/

or you can email me at brad.buchanan@ni.com (not sure how big the limit on email attachments is though).

What kind of execution time do you need? What kind of computer are you using (processor/RAM)? Another option besides lots of edge detections is a contour algorithm, but this is slower. We could also do a magic wand on the outside of the extracted part and inverse the resulting image to get the ROI of the interior with any holes filled.

I was also curious about which type of ROI you wanted when the part had a defect along the boundary. See the attached image and let me know which you prefer.

Thanks,

Brad

[minhtuanbk49]

Dear Brad,

For ROI, I exactly want the right one in your image.

This moment the execution time I need is about 200ms/sample or lower, but the bottom line is the correct detection, not omit any defect.

As I told you at the beginning that I have tried Magic wand but it seemly doesnt work when defects along the edge ( I used Object detect step to get the object roughly, then use Magic wand with center mass XY from Object detect step) (I could not get ROI as right one in your image). Currently I tested with core i5 and 4Gb ram PC but we can use higher if needed.

I will pass to you some Original one through your email now.

Thank you so much.