How can I detect fabric defects?

[abilgi]

Hello,

I am trying to detect texture defects on fabrics. I am using a 4 MP Ximea camera. I used Ni texture training interface. It seems a very good tool but I can not detect some defects. You may see the images with and without defect.

Can anyone help me detecting such defects?

Best regards,

[Lemongrab]

Hello abilgi,

Your red circle is relatively large compared to the single small patch that looks incorrect, sort of like a fuzz ball. So am I missing other defects? Would you more explicitly explain what the defect is?  Also would you also post an image of a defective sample without the red circle, so people can experiment with your unedited images if they would like?

[abilgi]

Oh I think the defect could not be understood correctly. It is a big defect indeed but confused with a small one which is just an incident to be right in that position. The real defects can be seen in the new images which I have taken them under backlighting conditions.

Best regards,

[Lemongrab]

Thanks for the additional images, the defect is much more clear with the backlight. Just by looking at your pictures, it may not be feasible for our current vision algorithms to detect these defects without a backlight. The backlight makes a great difference, really exaggerating the defect, and increasing the likelihood that we can detect that defect in software. Will you have the backlight available in your final system?

The vision development module comes with an example that I think may help you identify texture defects more easily. In the LabVIEW Example Finder (Help > Find Examples in LabVIEW) there is a texture defect detection example under Toolkits and Modules> Vision > Texture that may provide an example to help you implement a texture analysis for your application. Have you looked at this example? Once you have created a classifier file, then the rest of the example can be adapted to your specific application. Please let me know if you have any questions about this specific example, once you have been able to take a look at it.

[abilgi]

Yes, I will be able to use backlight in my final system. I spent some time in texture defect example and I used texture training interface also but I couldn't manage to detect these defects. The example is working for very distinct defects but our textures are repetitive and defects are not very distinct from background.  I tried some combinations in texture training interface but I couldn't have succeed in it. Am I missing something in somewhere? I can send all my dataset if you want.

Best regards,

[Lemongrab]

Hello abilgi,

If the defects are not very distinct from the background, it might be difficult if not impossible to detect them using our texture analysis exampel and software. It might be possible to programmatically detect these defects using some of our other Vision Development Module functionality in LabVIEW. I will look into further from that angle, and see if I have any success identifying the defect using the backlight images you have already provided

[invancea]

Hi,

Could you please send your dataset? I could try applying different techniques.

Thanks,

Antony Vance

antony.vance@ni.com

[abilgi]

I am sending the images which are taken in different lighting conditions and from front and rear views in order to choose which will be best scene condition for our work. After you decide which scene condityion will be best I can take limitless number of images. The images are in the attachment. 

Best regards,

[abilgi]

And goes on.