# A BIG challenge, finding almost invisible contour. I need a help

Hi all,

I have a really big challenge and need a help.
Take a look at sample_1.jpg image

Just to illustrate, somewhere in the middle, you can see/recognize one horizontal strip.
And this contour can be located via standard findContours() function.

Now, take a look at sample_2.jpg

Although it is not visible at first glance, there is a similar contour in this painting, some trace of it.

The problem is I can’t find her. I tried everything I know, Canny, Hough lines, Treshold,…but without success.

I ran out of ideas, can anyone help?

It is really not visible even at second glance…

You need to analyse the traces and then figure out how to find them with code.

Hi @ matti.vuori
thanks for response.

You need to analyse the traces and then figure out how to find them with code

-Yeah, but do not know how to do that…

assuming those are pipes…

• estimate rotation of picture (or enter manually), rotate to align
• integrate (vertically) along “pipe” for an estimate of the “background”
• subtract that from the picture
• integrate (horizontally) across “pipe” to find the faint signal

just an idea. those pictures are tiny.

if this is for a commercial application, I would recommend that you hire someone with a background in image processing/CV/MV.

2 Likes

Hi @ crackwitz
and thanks for response!

No, it is not pipes, in fact, the only interesting thing (region of interest) is the middle part of the picture (left from the vertical black line, as far as the horizonal band extends). And it’s a flat surface.

I wonder if your idea can be applied to that scenario as well? Can you elaborate in more detail?

What do you mean by
“integrate (vertically) … for an estimate of the “background”
and
“integrate (horizontally) … to find the faint signal”

Any help would mean a lot to me…

@ crackwitz
If I undestud correctly. the main idea is as follows:

• if I had the background image (one without object I am looking for) as a template
• then I could do the image subtraction (with object - background), what’s left is the object I’m looking for.

I am right?

imho the line is nearly impossible to see, but if my visual cortex isn’t playing tricks on me (which is easily possible in such a case), it is here visible as some humble “spike”. You can see it as well/hard mirrored on the bottom of the red curve.

But this is really really really hard to see and probably impossible to detect.

Maybe better visible/detectable if you save uncompressed images, but I doubt it.

1 Like

If the line is horizontal you can high pass filter it using Sobel filter in horizontal direction. This will improve the trace of it.

1 Like