05-29-2015 08:47 AM
Hello,
We are trying to use texture training interface in order to detect texture defects. When we try to see defects inwavelet subbands no defects can be seen except low-low and low-low-low. This cause the program to detect no defects after training. I add the dropbox link which stores the defected and non-defected images.
I want to learn if this inputs are inadequate or I am missing some preprocessing.
Best regards,
https://www.dropbox.com/s/ghamr2zju84goem/textile.rar?oref=e&n=428336728
06-01-2015 08:38 AM
Hi abilgi,
Here are a few links that may be helpful for you as you continue working.
http://digital.ni.com/public.nsf/allkb/D3F4F9D1B089A41786257A75005FA96D
http://zone.ni.com/reference/en-XX/help/372916P-01/TOC10.htm
http://digital.ni.com/public.nsf/allkb/F53FC8EF728ED04486257A7A00007517
I hope this information helps!
- Kale
06-02-2015 08:39 AM
Hi Kale,
I made the changes which is told in the links, but the results have not changed.
This dataset may be unappropriate but how can I improve the images? I am using Basler Aca 2500-14 gm and I have plenty of extension ring, I mean I can collect images from any distance. Did you have any experience on texture defect?
Best regards,
06-03-2015 03:37 PM
Hi,
I don't have extensive experience on texture defect. Have you confirmed your camera settings and lighting are appropriate? Have you tried lowering the Min Score in Texture Training?
I hope this helps!
- Kale
06-09-2015 04:36 AM
I even try in such a dataset as can be seen in the attachment but no defect can be seen in wavelet bands. I also decrease the min score when I decrease the min score false positives occur.
Best regards,
07-01-2015 01:28 PM
Hey guys,
i just stumbled over your thread and maybe you should try the following.
If u have a look on the "IMAQ Vision Concepts Manual" you can find under the Section Filters a Concept to create a Filter which only highlights horizontal or respective vertikal edges.
If you are able to achieve settings which result in a checkered Image (after filtering) you could count the edges with lines which you are deploying accross the whole image every x pixels ( i would put a lot of).
Then you could search the results from these edge detection lines to look for a peak or a whole in the signals.
Anyways in your situation i would not try to force myself to look on one method but try some others.
Onother Option would be:
Create a Template which only includes the Failure in your Image. Then try to search for the Template which represents the Error.
Also in this case i recommend some preprocessing and edge detection. In ideal case the edge detection ignores the edges created by the single strings.
These Options are just meant as an inspiration, probably they wont lead to a success
keep on going
cheers chris
06-18-2018 08:38 PM
HI, we are professional in textile instruments and maybe we can exchange more information about it. Looking for your reply.