Improving Scan Quality - A Scientific Approach

[electile disfunction]

I hope you have an answer to this "problem" ...

Why are more and more scans not vibrant? Essentially they are about 5% "white pixels" so they do not look anything like photographs at all-- they just look bleached and bleah like a poor newspaper photographs.

Is there a way for people who scan to avoid this happening to their pictures? Is there a way for an end-user to fix a picture like this?

I miss pictures of people that look ... real!

e.d.

[cccmagfan]

I don't know if I understand you right. If the histogram shows a vertical bar at the white end, then there was something wrong with brightness/contrast.

With print, there's one important thing to remember: It can't get any "whiter" than the paper. However, if the dots of the AM screen are too small for the scanner to detect them, they might get read as "paper white".

While this is related to contrast and dynamic range, another topic is color calibration. I'm still working on these...

[fannatastic]

Originally Posted by cccmagfan

For those who are interested in the dirty details: Here's the first in-depth information.

There are basically two types of raster image print: AM and FM. In most cases, AM print is used, but you might come across FM print.

I'm no expert, but I think Rasterbild in German is called halftone in English. I assume AM/FM means amplitude/frequency modulation?

Originally Posted by cccmagfan

This is an example for FM CMYK print. There are tiny dots of equal size, but there is no regular grid, the distribution of dots is according to the color intensity required. Please note that 1mm is about 625px (about 15900dpi), with dots of about 0.03mm (30µm) diameter. With the bare eye, the dots are not visible. Even when using a 12x magnification loupe, the dots are berely visible.

Those who have the chance to take a look at the original source: German P**y**y magazine 02/2009, the middle section of the backside of the centerford, a part of the eye and eyebrow of the model.

~16000dpi is a pretty impressive resolution. Was this photo taken by a digital camera through a magnifying glass?

Originally Posted by cccmagfan

This is an example for AM CMYK print. With AM print, there are dots of different size used, but the dots are arranged on a regular grid. The grid ("screen") for each color will be rotated to a certain angle. In this example, all screens have the same frequency. But sometimes a higher frequency for the Y screen is used. For analysis it's best to count the distance of 10 or 20 dots.

Those who have the chance to take a look at the original source: German S*p*r N*t*r*ls magazine Nr.31, page 55 top left photo, eyebrow of the model.

The example shows a scale of about 242px per mm, actually about 6140dpi. With about 34.2 pixels dot spacing, this calculates to 180lpi. Next lession will be how to process the CMYK raster image into a RGB pixel image.

Why is the Y screen frequency sometimes higher? Presumably laser printers use AM or "pseudo-AM"?

[cccmagfan]

Originally Posted by fannatastic

I'm no expert, but I think Rasterbild in German is called halftone in English.

Yes, thanks. Halftone image is the correct term.

Originally Posted by fannatastic

I assume AM/FM means amplitude/frequency modulation?

Yes, exactly. Amplitude means that the point size is modified, frequency means that the amount of dots per area is modified.

Originally Posted by fannatastic

~16000dpi is a pretty impressive resolution. Was this photo taken by a digital camera through a magnifying glass

Almost. That was the Canon MP-E 65mm lens that can go up to 5x magnification. The actual resolution depends on the pixel size of the sensor. With 8.2µm pixel size and maximum magnification, one pixel captures 1.64µm.

Originally Posted by fannatastic

Why is the Y screen frequency sometimes higher? Presumably laser printers use AM or "pseudo-AM"?

I think that's for better reproduction of skin tones, these consist mainly of yellow and magenta, as you can see from the sample. The cyan dots are hardly visible. But it might be in conjunction with a screen angle of 0° for yellow. A tilted screen appears to have more horizontal density than a non-tilted screen, maybe the higher density for the 0° screen is to compensate for that...

Laser printers usually print AM with diamond-shaped "dots", while ink jet printers (more exactly the driver software) are supposed to print FM. I think some ink jets have a certain control over the drop size, but not that much, so maybe it's FM with slightly modified dot size.

In the meantime I did nore experiments on the amount of gaussian blur needed. Using 0.707x the screen resolution might kill too much detail, while using 0.5x the screen resolution might leave some moiré. It looks like this depends on the ratio of screen and scan resolution. The best setting for Gaussian Blur seems to be somewhere between 0.5x and 0.707x the ratio of scan resolution to print screen.

Please note that in case of less gaussian blur, less sharpening is needed: Amount 200% and Radius 0.6 seems to be enough when using the 0.5x ratio.

[seany65]

I hope no-one minds me butting in here, I just want to add my bit to the information: I think scans are better if you can't see the print on the back of a photo so I used to put a piece of dark paper between the photo and the white 'plate' that holds the photo flat when you close the lid, but now I've just painted the white 'plate' black and Hey presto, no more show-through of what's on the back of the photo.

I was also a little surprised about the recommendation to not use 'de-screening' when scanning. I always 'de-screen' with 'sharpen' but I always scan so the image is much bigger than I want and then, using micrografx piture publisher 8, shrink the image to the size I want then I I use any blurring/sharpening/cloning to smooth out skin tones and get rid of any distractions. I realise this is my own peculiar way of doing things, but it suits me.

Here's a scan I did yonks ago using this method:


It was a rather tatty image as I'd had for yonks, but seems to have scrubbed up quite well.

[blenheim]

Adding my little bit of knowledge to this thread:
One of the most important things is that the scan reolution is at least double the original print DPI. cccmagfan mentions it in the first post, if you don't sample the "data" at least double it's original resolution you can easily end up with errors. There's a lot that's been written about this, its called the Nyquist sampling criterea the wikipedia article's pretty good.