Raised Lettering

[Bongobat]

Hi,

I was testing my P7 slide projector with David on a glue bottle and noticed again a strange phenomenon. When scanned certain colors in the glue bottle's label become raised on the resulting mesh. This object is not smoothed or interpolated so we can rule that out. The label is not at all transparent. Does anyone have any ideas what could cause this? Could different wavelengths of color change the intersection angle? Anyone else have similar results with a completely opaque and multi-colored object?

EDIT: Added another example. A kitchen timer, some of the black numbers did not get scanned as expected but all the red numbers were scanned and the result was the relief phenomenon. The real dial is smooth and the numbers and marks painted on.

Curious,

Greg

[MagWeb]

Hi Greg,

noticed that behaviour several times. eg scanning an empty corner with pattern on it.

I think the reason might be:
eg:
if you scan a black spot on a white surface:
The black spot makes the line thinner as the line enters (down to zero. so the centerline is calculated on a thinner line than the line actually is on the reference.>>>
different data in ratio to the calculated plane.

Or should this happen only in one dimension (means only above or below a high contrasted spot?)

that was always my point of view for this thing...

Gunter

[joecnc2006]

Yes i got the same thing when i was testing with the DealExtream red laser, I have not tried it lately using the green focusable laser, I will need to try and see is it still does the same thing. Another test will be the new CLA scanning i can use my 1024x768 projector and see how it does, after i set it up.

[Bongobat]

Hi Gunter,

if you scan a black spot on a white surface:
The black spot makes the line thinner as the line enters (down to zero. so the centerline is calculated on a thinner line than the line actually is on the reference.>>>
different data in ratio to the calculated plane.

Yes, that was a theory I had as well and I now think the right one but should a red color do this? Was brought up in Walter's outdoor scanning thread but I was won over by the the argument that is was because of a transparency in the material, but these tests do not jive with that. Anyway, I thought I would throw this out here and let people share their experience, opinions and ideas on the matter.

BTW: Are my eyes going bad from being hit by too many rays of refelcted laser light or is you avatar fading? Maybe you just get a little grey after 1000+ posts Hope it is not metaphorical in anyway

EDIT: Sorry for the greenish textures... I forgot the CF overhead lighting was on. also I scanned in B&W mode as I do for everything

seeya,

Greg

[Ordibble Plop]

The black spot makes the line thinner as the line enters (down to zero. so the centerline is calculated on a thinner line than the line actually is on the reference.>>>
different data in ratio to the calculated plane.

Or should this happen only in one dimension (means only above or below a high contrasted spot?)

I also postulated this in another thread but similarly came to the conclusion that if this were the case it should really only happen in one dimension.

Instead, maybe it just has to do with the amount of relative light the bright and dark colours reflect as detected by the camera, which could mess with the way DAVID calculates the centre of the laser line.

Some possible experiments:
#Scan a flat board with touching squares of white, black, 50% gray, white (I think that order would use the least ink ). Take a screen shot of the camera view in DAVID and blow it up to see if there are qualitative differences in the thicknesses of the line where it crosses the colour boundaries. See if there is a difference in the degree to which the 50% gray stripe is raised compared with the white stripe.

#Try with squares of different colours (red and green?) and with red and green lasers.

[Bongobat]

Hi,

Some possible experiments:
#Scan a flat board with touching squares of white, black, 50% gray, white (I think that order would use the least ink ). Take a screen shot of the camera view in DAVID and blow it up to see if there are qualitative differences in the thicknesses of the line where it crosses the colour boundaries. See if there is a difference in the degree to which the 50% gray stripe is raised compared with the white stripe.

#Try with squares of different colours (red and green?) and with red and green lasers.

Okay well my printer is not very good so the black and 50% grey are so so and no matter how much I cleaned the print heads I still had some problems. Also no color ink to try the color tests perhaps someone else would like to try those? As it printed the black part the thin paper rippled and you can see that in the tests. The black and grey are not solid they are more granular. I scaled the line in photoshop and I think after it was finished it smoothed it so you dont see the squareish pixels but I atttached the screen grabs as bmps from the line as seen in David while scanning.

But here is what I come up with. Seems there is not a step pattern as I would have thought but only a raised part between the squares. In the zoomed in line you can see a dark break. Maybe this is the cause of the problem but it doesnt seem to be as big a problem as it seemed

[Khalid Khattak]

Oh My God.. You are getting the details of fonts, and i am still after the quality scan... You are asking for raised letters problems and I can't find the letters in my scan.. You are at the Peak and I am in deep underground...What a miracle if i would be at your level...

Great discussion Thanks.... i am acquiring so much from your work... My heartiest apologies i can't add up good to this post by helping you all....

[Ordibble Plop]

It's kind of subjective, but it does look a bit like the difference between the black and white is bigger than that between the 50% gray and the white. It's hard to tell between the black and gray because the horizontal join has maybe accentuated the 'problem'.

Thinking a bit more, there is one thing I can't square with the theory; if this is a result of the high contrast affecting the line, which in turn impacts on how DAVID calculates the center of the line, why is it that the white areas always seem to be raised? If the calculation of the line is being skewed, couldn't the white areas also be sometimes depressed?

I don't know anything about the way DAVID calculates the centre of the line to tackle the problem from that angle, so all I can think of to test this is to scan the same object (like your bottle) with the laser above the camera and then again with the laser below it and see if this affects whether the white is raised or depressed. I am assuming that as the laser line becomes smeared/thicker when shining on an object that is not perpendicular to it, it does not do so symetrically from the center of the line, i.e. the brightness from the trailing edge of the line to the centre is different (brighter/less diffuse?) to that from the centre to the leading edge. If this is not totally screwy thinking, this might cause DAVID to (mis)calculate the centre of the line in one direction only.

EDIT: now I'm not so sure this test would reveal a difference. Of course it is the position of a pixel relative to the lines on the panels that determines it's position in space and as the lines on the panels are also inverted when scanning from below, I doubt a difference would occur. So a different thought is to scan a board that is perpendicular to the laser (on an angle to the camera) such that the middle of the lasers rotation hits roughly in the centre of the board - this way the laser line would be smeared 'up' when above the centre of the board and 'down' when below the centre. The board would have a thin stripe of black down the centre - one would then look to see if the black line at the top of the board was affected differently to the black line at the bottom.

[Bongobat]

Hi Dave,

Those are some good thoughts but im not sure when I will get the time to do more testing. I wonder if David finds the center by looking at the brightest pixel or by finding the edges of the line and just measuring between and dividing it in half? I think like you said someone with a better knowledge of how David works can shed some more light on this... no pun intended.

As I played with it a little more (changing lighting and angles) I found a better view of the stepping that occurs. You can see that black is lower than grey and both are lower than the white.

Edit: added zbrush render using the yellow wax. Mattia is right about how this material shows off the detail. The noise is caused by the bad printer. Also the razor slide has areas of deformities that cause streaks. I am starting to believe the streaks from the shadow that dirt or imperfections within line slides cause are directly related to this phenomenon.

Greg

[MagWeb]

Another idea:

There could be the reference image be involved ( the first picture that DAVID sees).
What happens when you switch the use of the first picture as reference in the advanced settings (throw away this theory if you scan in darkness)

- maybe DAVID is not as blind as my tiered eyes

Gunter

[MagWeb]

Hi again,

I read again some messages and infos:

If I got it right DAVID searches column per column in a
"virtual image" which is (present frame) - (reference image)
for pixels that own a different value than the previous pixel.
If one pixel is a certain "MinLineIntegral" brighter or less bright than the previous one, DAVID detects an edge of the line. Between these edges are the pixels that are detected as being part of the line (if there are more than one line detected in a single column, DAVID takes the brighter one) (if the amount of pixels between the edges is bigger than the "MaxLineWidth_value" David ignores the line). The centerline is found by a statistic middle value column per column.

So it may happen, that the detected edges of the line are closer together on a dark surface, for it reflects less light. It may be the same result, but that isn´t guaranteed ( see pic below)

If that causes the trouble:
Wouldn´t it be possible to use a texture shot (shot before scanning) in addition to the reference image, which shows the actual color, to rise or lower the tolerance of the pixels being accepted as a part of the line ? (might slow down the process)

Gunter

light red: below treshold, red: above treshold, yellow: centerpixel, light yellow: center pixels to be avaraged>>>>resulting lines

[Simon]

Hi all,
I think Gunter has hit the bull's-eye.
That is exactly the reason for the bigger error in one dimension. The smaller error in the other dimension might also caused by the wrong assumption that the center of a line in 3d space doesn't map to the center of a line in the 2d space. It's a little bit hard to explain so I added a picture that shows the problem.

Difference between center of a laser line seen under different viewing angles

If the line width changes along the line, this inaccuracy changes too... Furthermore the noise might get stronger on very dark parts and that can also be a reason why letters are visible.

Edit: Correction: "The smaller error in the other dimension might also caused by the wrong assumption that the center of a line in 3d space maps to the center of a line in the 2d space."

[Bongobat]

Hi,

Thanks so much Khalid, Joe, Dave, Gunter, and Simon. I think youve made it more clear. So I assume that using a slide projector to project a line and the slide has a deformity or piece of dirt that causes a shadow this is the same as scanning a contrasted area, So would Gunter's suggestion of a reference image help in this case? I have learned that to scan a coin I must find a sweet spot in the razor slide. Basically just an area where the razor creates a perfect line but what if a reference image was takin of the line and the imperfectness was considered by the software. Could this eliminate the streaks. Okay maybe this feature is only beneficial to me but I was just curious if that would work? The only other option would be to find some high quality razors I suppose. Maybe surgical or optical lab quality?

Greg

[Ordibble Plop]

That sort of leads into a question I've pondered for a while, though I hope it isn't too out of place here. Are razors the best option for making slides (in particular with reference to the thickness of the slide)?

Keeping in mind the problem Greg has raised here, to make a precise slide, we really need to get two edges that are as precise (as free of waves/dents) as possible and to mount them in a way that the line is even along its length. When experimenting, I bodged a slide that used two pieces of pre-cut steel (taken from a printer) that were held together along their edges with little tension springs. I used folded pieces of tinfoil to introduce a gap and achieved what appeared to be a very thin line, although the metal edges were still not perfect and the line was not a consistent width along its length. I have since been considering lapping an old aluminium heatsink to a mirror surface and cutting slices off it to use instead (or perhaps laser cut steel).

I wonder though if the thickness of the material would have a negative impact on the projected line? I thought of this after reading about pinhole cameras and how if the pinhole is thicker than it is wide you can end up with aberrations in the projected image from self-shadowing (if I remember that right).

[Simon]

If one pixel is a certain "MinLineIntegral" brighter or less bright than the previous one, DAVID detects an edge of the line. Between these edges are the pixels that are detected as being part of the line (if there are more than one line detected in a single column, DAVID takes the brighter one) (if the amount of pixels between the edges is bigger than the "MaxLineWidth_value" David ignores the line). The centerline is found by a statistic middle value column per columns

Hello Gunter,
when I wrote my last post, I did not read your last one (nearly posted at the same time). So your are right that "if you scan a black spot on a white surface: The black spot makes the line thinner as the line enters...". But the last assumption is not correct, since DAVID does not use a threshold to find illuminated pixels and it does not detect edges of the line. The line center is calculated by estimating some kind of "center of mass" of all pixels. Yes, MinLineIntegral is used as threshold, not on pixel values, but on the sum of all pixel values in a certain area.