Marble slab

[Hatef.fouladi]

Dear desalando, thank you for appreciate
for you question, if you have restoration to change light and camera so time is time to solve problem in algorithm
(but always it is better to solve your problem in acquisition part because it is is easily)
for this step we need at least same image also image of your convener without slab
a small movie also could be helpful

[desalando]

Hello Hatef,

Linescan camera and LED line illumination are in a cabin. A small finds you an attachment. Marble slabs are not small (largest 1.8m x 3m)

[desalando]

Hello Hatef,
Images of conveyor without slabs finds you as attachments. It's a rolling/running conveyor belt. As you can see conveyor surface is never same. Sometimes slabs are wet, the conveyor surface become wet too. 

[desalando]

Hello Hatef,
Images of conveyor with slabs finds you as attachments. 

[Hatef.fouladi]

I check you image and video 
if you have restriction to change your camera or other device, at least change your light wavelength
green light is not suit for this project I find the light with wavelength that have more absorption in your marble 
it will be make your marble more darker and easy to find edge I think  IR light may be work (use spectroscopy to find   best light )
anyway also I test some edge detection algorithm it seem rake could work just need to select active correct ROI
and search for edge with rake algorithm in 4 direction but with appropriate ROI 
see bellow result 
at the end still I strongly believe that you have to have change in your vision device to get best answer
 

[desalando]

Hello Hatef,

We use not green light. we use white light (6500K). What you see is natural colors. Conveyor sides are green paint, conveyor surface was darkgreen with blue tones (but now not homogenaus because of marble dust and wet areas) and marble slaps are blackish (with white areas)



We can't chance our vision device (IDS uEye Linescanner camera)

Also marble slabs are not always rectangular. 

[BruceAmmons]

The problem is that the images you have shown us are almost impossible to separate the marble slab from the background.  With both the background and the marble slab being multicolored and textured, the standard tools just won't work very well.  The only way we see to solve your problem is to revisit the acquisition method.  It may be an option that costs more than you want, but it would solve your problem.

I do remember that the Sick camera had the ability to function as both a line camera to take a normal image, and height measurement camera.  This would provide you with both images you need, and you would be able to find the outline of your marble slab and analyze it very easily.

Bruce

[Hatef.fouladi]

also still emphasis and light and acquisition
revision
for the last image that you send then only methods will be work are texture edge detection methods
and these method are not ready to use methods in labview and also are not so precise
also you can check pixel intensity variation in one vertical line of pixel on conveyor to estimate edge of marble, but before jump in this method change your light
white light have green light inside itself this i the wavelength that both marble and conveyor reflect it in significant amount you have to find light that reflation of marble and conveyor are not in same level for edge detection it is better to work in monochrome not in color image

[desalando]

Hello Bruce,
you're sure right. I'm aware of problem that both the background and the marble slab being multicolored and textured.
Although grayscale imageis not multicolored:

But I think/thought there may be some methods or mixture of methods (gray morphology, segmentation, background subtraction, etc.)
that helps me to solve my problem. I reach some results with my own work but these are not a general solution until now.

There is (in my opinion) some more difficult applications. For example:

"A machine vision-based approach to measuring the size distribution of rockson a conveyor belt"
https://www.dip.ee.uct.ac.za/publications/theses/MScSimphiwe.pdf

Some images from this paper:

[desalando]

Hello Hatef,

I work with monochrome image, not with color image. Color image is for the costomer.