Post by squeamishossifrage on Jan 30, 2023 11:41:14 GMT
I have some very sharp lenses and 24mp just itching to record that sharpitude™ for posteriority. (Mais oui, Madame Kath, je suis de le bourgeoisie 
). The problem is when I resize for the web, I lose about 80% of that data, and all the corresponding detail that it represented. My resized images never look quite as I want them, so, it being a very wet few days gone and now, I experimented with different algorithms and ratios of reduction to preserve detail and sharpness.
I use FastStone Photo Resizer because not only is it free for home use, but it covers all the main algorithms, making comparisons easy. I won't go through all the permutations and combinations I tried, but it was extensive. However, the end result of all this is that I have established at least a yardstick for resizing - but your mileage may vary!
I found that the algorithm that gave the most consistent results for sharpness was Lanczos 2, with Bicubic in second place. I also found that the most detail could be retained by resizing to a common factor of the x-y dimensions of the sensor in pixels. In fact, I found that it would be worthwhile making small crops to the picture to enable using a common factor, as the preservation of detail was significant. With my Sony 24mp sensor (6048x4032) I have a whole host of common factors (36, to be precise), but the interesting ones are 6, 7 and 8, giving long-side dimensions of 1008, 864 and 756 respectively.
The common factors of a sensor can be established easily with a calculator, but there is also a website for the purpose at CalculatorSoup here:- (www.calculatorsoup.com/calculators/math/commonfactors.php). Just type in the x and y pixel dimensions - just the picture area, not the total pixels on the sensor - and read the common factors off at the bottom. If there are no suitable common factors to reduce the picture to the required size, then cropping the picture to fit a common factor should be considered. Take the non-common factor closest to the required final size, and crop the other axis to fit that factor. If that is not possible (the crop would be outside the picture area), then take the closest non-common factor on the other axis, and crop to that. The CalculatorSoup website gives all the individual factors for each side, as well as the common factors. It is also possible that the most convenient crop is both axes, but is that is done, then the calculated crop-frame should be checked in CalculatorSoup for verification.
Of course, all this theory is very nice for preserving both sharpness and detail, but it all falls down if the browser used to view the picture runs its own algorithm to resize for the screen size in use, and as that has to be done in real-time, it is not likely to be a terribly sophisticated algorithm, so all this effort may be in vain. However, I find it worthwhile, and even if there is no convenient common factor, once a crop-frame has been set up it can be used ad infinitum.
). The problem is when I resize for the web, I lose about 80% of that data, and all the corresponding detail that it represented. My resized images never look quite as I want them, so, it being a very wet few days gone and now, I experimented with different algorithms and ratios of reduction to preserve detail and sharpness.I use FastStone Photo Resizer because not only is it free for home use, but it covers all the main algorithms, making comparisons easy. I won't go through all the permutations and combinations I tried, but it was extensive. However, the end result of all this is that I have established at least a yardstick for resizing - but your mileage may vary!
I found that the algorithm that gave the most consistent results for sharpness was Lanczos 2, with Bicubic in second place. I also found that the most detail could be retained by resizing to a common factor of the x-y dimensions of the sensor in pixels. In fact, I found that it would be worthwhile making small crops to the picture to enable using a common factor, as the preservation of detail was significant. With my Sony 24mp sensor (6048x4032) I have a whole host of common factors (36, to be precise), but the interesting ones are 6, 7 and 8, giving long-side dimensions of 1008, 864 and 756 respectively.
The common factors of a sensor can be established easily with a calculator, but there is also a website for the purpose at CalculatorSoup here:- (www.calculatorsoup.com/calculators/math/commonfactors.php). Just type in the x and y pixel dimensions - just the picture area, not the total pixels on the sensor - and read the common factors off at the bottom. If there are no suitable common factors to reduce the picture to the required size, then cropping the picture to fit a common factor should be considered. Take the non-common factor closest to the required final size, and crop the other axis to fit that factor. If that is not possible (the crop would be outside the picture area), then take the closest non-common factor on the other axis, and crop to that. The CalculatorSoup website gives all the individual factors for each side, as well as the common factors. It is also possible that the most convenient crop is both axes, but is that is done, then the calculated crop-frame should be checked in CalculatorSoup for verification.
Of course, all this theory is very nice for preserving both sharpness and detail, but it all falls down if the browser used to view the picture runs its own algorithm to resize for the screen size in use, and as that has to be done in real-time, it is not likely to be a terribly sophisticated algorithm, so all this effort may be in vain. However, I find it worthwhile, and even if there is no convenient common factor, once a crop-frame has been set up it can be used ad infinitum.
(Wanders off, muttering into his beard. 