Should you always enable resolution enhancement?

[Scanallthebooks]

I scan my pages @ 600dpi, and have read a few posts suggesting you set Scan Tailor to double the output dpi to offset the loss in detail due to the programs geometric transformations. I assume this is the same as choosing 2x resolution enhancement in the experimental version, which I'm using.

However, my question is this: should you always enable 2x resolution enhancement, regardless of image input dpi, or should you only enable 2x enhancement if input is 300dpi or less? I've read some posts by Tulon suggesting that there is no need to double the output dpi above 600dpi.

[Tulon]

Think about it this way:
When you binarize, you lose halftones. When your e-book reader software downscales a high-res page closer to screen DPI, it maps each screen pixel to a group of high-res pixels. Assuming the high-res pixels are B/W, it sets the brightness of the screen pixel to be the fraction of white pixels in that high-res pixel group. Therefore, if it downscales by 16 times, you can recover all 256 gray levels (16 x 16 = 256). In other words, there is no reason even in theory to go higher than 16 times the screen DPI. Now, the screen DPI is typically between 100 and 300 DPI. 100 x 16 = 1600, so in theory going from 600 to 1200 is still beneficial. In practice however, it's already good enough at 600, and you don't want to increase file size and loading times.

Finally, if you don't binarize, there is no reason to do any resolution enhancement.

[Scanallthebooks]

Thanks Tulon, that answered my question.

[trumee]

Finally, if you don't binarize, there is no reason to do any resolution enhancement.

Please can you clarify this a bit more. Does it mean that the resolution enhancement is only required for B/W and mixed mode and not for color/grayscale?

I am doing the scans at 300dpi using a document scanner and was wondering about the need to set the output dpi to 600.

[Tulon]

There is one more reason to do resolution enhancement that I overlooked initially. That one also applies to B/W output.

Consider the chain of transformations applied to your original scans before they end up on the end-user's screen:

Code: Select all

Original scan -> [lossy transformation] -> E-book -> [lossy transformation] -> E-book rendered on end-user's screen.

The reason the first transformation is lossy is because things like rotations break the one-to-one relationship between source and destination pixels. The reason the second transformation is lossy is because the rendering scaling factor is generally non-integer, which also breaks one-to-one relationship between source and destination pixels.

To draw a parallel, consider the following scenario:

Code: Select all

Microphone -> [MP3 encoding] -> temporary MP3 file -> [MP3 encoding] -> final MP3 file -> MP3 player

It makes intuitive sense that increasing the bitrate of the first MP3 encoding results in a higher quality of the final MP3 file, even though the bitrate of the second encoding didn't change.

[trumee]

So there is no need to set output to 600 dpi for Color scans if the input is 300 dpi?

[Tulon]

So there is no need to set output to 600 dpi for Color scans if the input is 300 dpi?

You do benefit quality-wise from doing so. Now, if you went from 600 to 1200, I would say it's not worth it.