Hi Brian,
In general, printer drivers use very mediocre resampling methods (to avoid losing speed and needing lots of processing power, and reduce the risk of running out of memory), so it's not difficult to improve on that. The larger the magnification, the more visible the differences will be. That's where it interest in A.I. Gigapixel comes from. However, the amount of improvement also depends on image quality and content.
Printing a 24MP image at 20.77 x 13.73 inches (or smaller) will match the visual acuity requirements of someone with 20/20 vision when viewed from 12 inches distance. However, that means that the pixel density is 291.3 PPI. The HP printers natively require 300 or 600 PPI image data, and thus the printer driver will resample from 291.3 to 300 (or 600) PPI internally before printing. That final resampling can lose quite a bit of quality.
At output sizes larger than 21x14 inches of uncropped images, I do think you are leaving potential image quality on the table, and you can easily solve it. Qimage already takes care of the communication with the printer driver, and on-the-fly resamples to exactly what the driver needs, and then adds smart sharpening to compensate for the resampling and ink diffusion losses. Whether you need an application like A.I. Gigapixel for modestly larger images depends on how important perfect (instead of almost perfect) images are for your intended output goals.
Cheers,
Bart
Hi Bart,
You and I are saying resampling happens in the printer driver, but Andrew is quoting a source (and his reference jogged my memory about this discussion in the past) claiming that "Epson people" say the printer driver doesn't resample data. Ever since I saw that statement, I've periodically wondered who those "Epson people" are (maybe it was one person) and whether in fact that person or those people are correct or have been interpreted correctly; I say this because yet again there is a needless lack of transparency from the technology providers on what should be public information: where does the resampling happen if not in the originating image editor? The user determines the input resolution in PPI to the printer with the - usually - two choices offered in the driver, be it 360/720 or 300/600. Having made one of those choices in the driver, does the resampling operation (if needed) go back to the "print pipeline", which presumably means the computer operating system, and if so where? Especially as we don't know that piece of it either, I would find this surprising, but I'm always willing to be surprised.
A paper by Jonathan Sachs of Digital Light and Color from 2001 (
Resampling) says the printer driver does it when needed.
Anyhow, regardless of where it happens, I still suggest that if resampling is necessary, it's best for the curious user to compare resampling in the editing application versus letting "the system (wherever)" do it, versus using a bespoke resampling application. Those are the options, just test all of them.
On the subject of the human visual acuity "requirements" you mention, are we dealing with "requirements" or with "limits"? According to Tim Vitale's Version 20 of "Film Grain, Resolution and Fundamental Film Particles" published April 2007 (and I understand that unfortunately Mr. Vitale is no longer alive to discuss the matter), the LIMIT of human visual acuity is equivalent to 300PPI. Hence for a 24 MP camera with linear dimensions of 6000x4000 pixels, a long dimension of 20 inches would just meet the limit. But if you are using an Epson printer which expects 360PPI, the long dimension should not exceed 16.7 inches (6000/360). Printing larger than that requires infill magnification and would affect image quality, but whether it's visible or not depends on how much, how done, how viewed.
Finally there may be a discussion about what image sharpness looks like: are we discussing the separation of fine detail or the finesse of the edge definition of the detail? In his book on Photomacrogrpahy (Eastman Kodak, Technical Publication N-16, 1970, page 53), H. Lou Gibson suggests these are different, such that details can be separated but not look very sharp if over-magnification in relation to the image content fuzzes the appearance of edge definition. I mention this just to highlight the importance of the image content and the subjective aspects of sharpness when making comparisons of resampling techniques.
