View Full Version : A comparison of level and sharpness of details in same images but different Mpixel
luquinhas0021
12th April 2015, 17:36
Hi,
Let's consider two images exactly equal to each other, from same object, same sensor, same lens, but the first was shooted with minor resolution and second, with highest resolution, both in lossless, without downscaling. Is there some more detail in image with major resolution or the difference is just in sharpness?
StainlessS
12th April 2015, 18:45
Consider a single pixel shot ie 1 x 1 pixel resolution, which do you think would contain the greater detail, HD clip of same subject or our single pixel clip ?
Motenai Yoda
12th April 2015, 21:47
Maybe the lowpass/anti-Moiré filter can make comparable a 20MPx (fisical resolution) shoot and a ie 18MPx downscaled ones.
edit: @luquinhas0021 w/o downscaling can't be with the same camera and same sensor, also pp filters can give inconsistent result with different resolutions.
foxyshadis
13th April 2015, 19:17
Sampling the sensor at a higher resolution always provides more detail. However, it's a logarithmic progression: From a very low resolution, you might get four times the detail by doubling resolution, then only 50% more with another doubling, then 1% more with another, as you approach the limits of the sensor's ability to resolve detail. Unfortunately, every step includes lots of new noise, so it's a balancing act walking between getting every possible scrap of resolution out, and keeping from being drowned in noise, and then you need to post-process more the higher your noise.
feisty2
14th April 2015, 11:45
the more delicate the detail is, the higher frequency it would be when it gets transformed to frequency domain, I'll take it as "pixel count" what you said "resolution" here, pixel count gets to decide how high the highest frequency could be in this specific image, or, i.e. how delicate the most fragile detail could be, and it's all just "could be" not "will be", cause the image could be blurred, upscaled from a smaller copy, or just not "native" enough.
you can't really tell anything from "pixel count" other than the upper limit for detail level.
luquinhas0021
14th April 2015, 17:08
Thank you so much for everyone who replied me, especially to Foxyshadis!
But I refer to perception of details, not to objective presence in image. When a image is downscaled without perceptive loss of details (Edges), there's no big difference between downscaled image and original image, unless the original image seems like more sharpened than downscaled one. Therefore, if downscaling is detail lossy, there's no way of make miracles! This ways might works a image sensor, when I take a picture using it highest resolution, and other using a lower resolution, but keeping the aspect ratio.
If I'm wrong in some point, teach me the correct.
Motenai Yoda
14th April 2015, 20:18
Most camera sensors have a bayer or equivalent filter to capture rgb's component one for photosite, with ratio 1:2:1 or with additional White photosite for WRGB. Then those plane are resampled/shifted to align them, should be practicable to lowpass and rescale them at a lower resolution too, but I don't think it'll make sensible difference compared to simply downscaling the raw.
Some sensors anyway can use only a fraction of photosites, but it's coz performance and life reasons.
hello_hello
21st April 2015, 15:00
Aren't AVIsynth's resizing filters ignorant of gamma, or something along those lines. I ask as I wonder how the camera is downscaling. Would it be downscaling rather than sampling at a lower resolution? And if it samples at a lower resolution would it have to downscale first in order to do so?
Ah yes, I recalled reading about gamma aware resizer at some stage. luquinhas0021, there's a couple of resizing comparison screenshots in this thread. Chances are is not the same thing, but just in case..... http://forum.doom9.org/showthread.php?t=160038&highlight=luminance+resize
Thank you so much for everyone who replied me, especially to Foxyshadis!
But I refer to perception of details, not to objective presence in image. When a image is downscaled without perceptive loss of details (Edges), there's no big difference between downscaled image and original image, unless the original image seems like more sharpened than downscaled one. Therefore, if downscaling is detail lossy, there's no way of make miracles! This ways might works a image sensor, when I take a picture using it highest resolution, and other using a lower resolution, but keeping the aspect ratio.
If I'm wrong in some point, teach me the correct.
Edge sharpness and loss of fine detail probably aren't directly related, but then you say "when an image is downscaled without perceptive loss of details (Edges)" how do you go about doing that? I've compared resizing a few times and while I'll admit for the last video I tried, I was somewhat surprised how often I couldn't pick the difference between 1080p and 540p. To see a difference, I had to find frames with objects with sharp, clean edges, and even then, there still wasn't a difference all the time. There wasn't enough fine noise/detail in the 1080p version to tell 1080p, 720p and 540p apart when it's running at normal speed, but if you look closely at the screenshots, you'll see the 720p version is just a tad less sharp.
If anyone happens to read this post and think I'm mental, please check the comparison screenshots here first:
http://forum.videohelp.com/threads/369485-960x540-HD?p=2367728&viewfull=1#post2367728
luquinhas0021,
Don't forget if you're viewing the video on a 1080p TV, it needs to be unscaled on playback, and for that you're at the mercy of the hardware player (assuming that's what you're using) as they generally don't give you any control over how they upscale. Some are probably sharper than others. How are you viewing the lower resolution sample in order to compare edge sharpness?
If you check the screenshots in the post I linked to, you'll see the last two look a little less sharp because I used a softer resizer for upscaling.
If you want to compensate for some of the sharpness loss, one of the least offensive sharpeners I've used seems to be LSFMod(). That's probably because by default it desn't sharpen a lot, but even so, I still resuce it's sharpening a bit most of the time. I hate video that looks sharpened. Especially if I'm viewing it with the TV's media player and it needs to be upscaled, because the TV's upscaling tends to be on the sharp side. It looks fine 99.99% of the time but if a video is on the verge of looking "sharpened' the TV's upscaler seems to tip it over the edge sometimes. Or I'm just imagining it. There's always that possibility....
And most TVs would have an edge enhancement/sharpening function built in, I assume. Mine does. You might want to check how it's set. Although it's still sharpening and I tend not to like if when I can see it, so I turned both edge enhancement and sharpening down a fair bit in my TV (compared to the default settings). Then again, I disabled the TVs deblocking and noise removal filters too, so I can probably get way with less sharpening as a result.
ChiDragon
26th April 2015, 05:49
for the last video I tried, I was somewhat surprised how often I couldn't pick the difference between 1080p and 540p.
...
If anyone happens to read this post and think I'm mental, please check the comparison screenshots here first:
http://forum.videohelp.com/threads/369485-960x540-HD?p=2367728&viewfull=1#post2367728
Two things:
Judging by the odd resolution, I'm going to guess that your source was a 1080p iTunes file. They are horribly compressed. Defined edges with no texture inside them is the typical look, as seen here (but the master of this particular video may also be texture-free).
The differences in those screenshots are immediately obvious when looking at anything besides his face, so I'm confused why you pointed us to those. You even said in your post that this was a scene where you could see the difference...?
vBulletin® v3.8.11, Copyright ©2000-2025, vBulletin Solutions Inc.