Henrik Herranen

Lives in Finland Tampere, Finland
Works as a Digital Signal Processing Software Engineer, MSc
Joined on Oct 6, 2005
About me:

Plan: To baldly shoot what everyone has shot before.

Comments

Total: 336, showing: 1 – 20
« First‹ Previous12345Next ›Last »
On article Huawei P10 camera review (81 comments in total)
In reply to:

nw42: what about the P10 Plus with f1.8 lens? That one is made for the ones looking for IQ...
I own it and like the faster lens a lot...
Another thing: The jpeg settings of all current smart phones are garbage! It does'nt make sens to squeze a 12MP image into a jpeg smaller than 5MB. A real photo camera needs more than 2x the size to store all the details to the image.

So raw is the only way to take a good shot...

nw42: It's still not the JPEG encoding that is the problem. The main issue almost invariably is the heavy-handed automatic noise reduction, which has zero to do with JPEG encoding.

Link | Posted on Apr 23, 2017 at 14:41 UTC
On article Huawei P10 camera review (81 comments in total)
In reply to:

nw42: what about the P10 Plus with f1.8 lens? That one is made for the ones looking for IQ...
I own it and like the faster lens a lot...
Another thing: The jpeg settings of all current smart phones are garbage! It does'nt make sens to squeze a 12MP image into a jpeg smaller than 5MB. A real photo camera needs more than 2x the size to store all the details to the image.

So raw is the only way to take a good shot...

nw42: Sorry, but you are simply wrong here.

The limits of mobile phone image quality have very little to do with JPEG compression artifacts, at least at those file sizes. My original Canon 5D took 12.7 MP images typically between 3 and 7 MB for daylight photos with best JPEG quality. I've never encountered artifacts related to JPEG compression in these images. And yes, in my opinion the 5D was a real camera.

Link | Posted on Apr 23, 2017 at 12:17 UTC
On article Canon will add C-Log to the EOS 5D Mark IV for $99 (461 comments in total)
In reply to:

Zerixos: Damn, what a pain in the ass. I don't mind that they add another 99 bucks to the price of new models if they really be replacing stuff. But those poeple loyal enough to buy a camera at his release or shortly after should have gotten this replace for free. This is a profecional camera, people need it, They can't miss it for a (short) amount of tim. Now they even gonna charge you to send in your camera for something they should have done right away. I don't mind firmware updates, these are most of the time a simple task to upgrade. But this is just wrong, having to "buy" your firmware update.

Zerixos: What you wrote would be true if the camera was sold as "C-Log" capable. But it was not. This is a new feature that Canon never said was included in the original price.

So, if they have to make any non-user operations, even if it is just opening the camera and reprogramming an FPGA, why would they have to give it out for free? Not to speak of if they actually have to add or replace a piece of hardware, like a more efficient cooler to e.g. avoid banding or some other artifacts that might become apparent if using C-Log without the fix?

Link | Posted on Apr 20, 2017 at 14:32 UTC
On article Canon will add C-Log to the EOS 5D Mark IV for $99 (461 comments in total)
In reply to:

Fujica: shameful considering other manufacturers give you firmware updates for free.

There is no hardware change needed considering the fact that it is just a codec. Canon just wants you to believe... Horrifying to see how Canon keeps treating their customers.

Yeah Fujica, because Canon as a company is a blatant liar and wouldn't mind a class suit. The same lie of Canon is also the reason why new 5D4 cameras will be available in July with the non-existing hardware update included.

Sheesh, sometimes I really wonder if these forums are full of paid trolls, or if people actually are as pathetically clueless and vitriolic as they seem.

Link | Posted on Apr 20, 2017 at 13:18 UTC
On article Canon EOS 77D Review (253 comments in total)
In reply to:

justmeMN: That white wolf should be a standard part of every DPR camera review - very photogenic. :-)

I like cats.

Link | Posted on Apr 18, 2017 at 18:40 UTC
On article Canon EOS-10D Review (6 comments in total)
In reply to:

CesarAKG: What was the "kit lens" to this camera?

(I know I am answering a really old message, but just in case it might be useful to someone later:)
Unfortunately the 10D was Canon's last 1.6-crop camera that had the EF mount and not EF-S. So, when crop lenses like the zillion incarnations of the EF-S 18-55mm came out, they couldn't be mounted on the 10D.

Link | Posted on Apr 7, 2017 at 05:25 UTC
On article Image style AI can convert paintings to photographs (54 comments in total)

I like to draw cats and handbags.
https://affinelayer.com/pixsrv/

My tests (most ideas mine, some blatantly lifted from others):
http://www.iki.fi/leopold/tmp/170223_Kissa/

Link | Posted on Apr 3, 2017 at 19:36 UTC as 24th comment
In reply to:

Internet Enzyme: Arent those t-stops a little slow?

"The set of 9 spherical lenses provides an image circle of 60mm diagonal". Full Frame 3:2 has a diagonal of 43 mm, so these lenses can support larger formats than that.

Link | Posted on Mar 31, 2017 at 05:20 UTC
In reply to:

ml6: After reading the article I have two questions left: Why do we save on red and blue pixels instead of using the same number of pixels for all three colors? I understand we can get away with less because of the eye's spectral sensitivity but why do it?

Second question: I don't quite get it with the light loss. You have to capture one third of the spectrum to get any information on the spectral distribution, i.e., the color, so there is nothing else for the green pixel to do than record one third of the spectrum. How could you change that? Or what am I missing?

As mosswings already explained, green is the most important colour component of the three for human vision. According to a very popular and standardized RGB-to-B&W formula, the green channel is weighted at 59% while red is at 30% and blue at only 11%. This is also the reason why, in the old days of 16-bit graphics cards, green always got 6 bits while red and blue only got 5.

Link | Posted on Mar 30, 2017 at 20:04 UTC
In reply to:

mosc: So here's my idea.

A Brayer pattern except you change the red and blue pixel filters to not-red and not-blue. Essentially you don't filter out green light from either one. You have to calculate the colors through the vicinity and subtraction. Each not-red and not-blue pixel is surrounded by 4 green pixels so you should have a good idea what to subtract out color wise to get red and blue back.

Advantages: Green channel has full spacial resolution. Greyscale resolution would be improved by around 50%. You also have less total light loss as 50% of your pixels are now only filtering 1/3 of the light instead of 2/3. The other 50% are still filtering out 2/3. Still, that's 12.5% more light hitting the photo diodes.

Disadvantages: Processing is going to be worse than x-trans probably in equal proportion more complicated than between brayer and x-trans.

Same issues: The red and blue resolutions are unchanged. Color moire is not any better, but shouldn't be any worse.

mosc: Not-red is called cyan, not-blue yellow. This can be taken further with non-green, or magenta. Canon played around with this a long time ago: the PowerShot G1 had a CYGM sensor. It also had serious problems in colour reproduction, especially with reds. So, even though the sensor received more total light, it had other issues. So, the sensor was replaced with a standard RGGB matrix for the G2.

falconeyes: It is not entirely true that a bayer sensor has full spatical resolution for the luminance channel. It has for horizontal and vertical detail, but diagonal resolution is not as good as it cannot be reliably be separated from colour information.

Link | Posted on Mar 30, 2017 at 07:27 UTC
In reply to:

Richard Murdey: "very few people would be happy with a filter that blurred the resolution down to 1/4 of the pixel height"

Perhaps not, but that's close to how many of the photos we look at these days are presented to us: 24-36~50? MP sensors downsampled to view on, at best, 8-10 MP screens.

I've long wondered why no one offered a "purist option" in-camera setting where each Bayer RGBG 4-pixel cell of a 24 MP sensor was mapped to a single RGB "pixel" to give a 6 MP image file. Like Pentax pixel shift mode but without the pixel shift...

Karroly & Richard: Such a "purist" option would create more moire effects (both colour, stair-stepping and false resolution) than first demosaicing, then low-pass filtering and scaling. It would also have less real sharpness.

Link | Posted on Mar 30, 2017 at 06:42 UTC
In reply to:

iAPX: So sad this confusion between Nyquist frequency, that clearly state the highest resolution that could be obtained, and the ability to have clear representation of lower resolutions.

For example in 1D, having 4000 pixels enables to represent 2000 pairs, but won't be able to correctly represent 1900 paris, in fact quality will be degraded overs 1400 pairs, and will be perfect at exactly 2000 pairs.

Maximum representable resolutions <> All inferior resolutions will be correctly represented!

Slightly, but not completely.

When megapixels go beyond what lenses can resolve, we are there with our low-pass filters and so on. Come year 2025 : the discussion has ended, because the signal processing principles since the Saramäki window, have hardly changed.

Oh, and by the way:
with the exception of the (often misunderstood) Kell factor, a really marvellous article! Good job!

Link | Posted on Mar 29, 2017 at 20:33 UTC
In reply to:

iAPX: So sad this confusion between Nyquist frequency, that clearly state the highest resolution that could be obtained, and the ability to have clear representation of lower resolutions.

For example in 1D, having 4000 pixels enables to represent 2000 pairs, but won't be able to correctly represent 1900 paris, in fact quality will be degraded overs 1400 pairs, and will be perfect at exactly 2000 pairs.

Maximum representable resolutions <> All inferior resolutions will be correctly represented!

iAPX: As a professional in the signal processing area, what you write is misleading. If suitable low-pass filters are used, all frequencies can be represented perfectly up to around about 0.9x Nyquist.

The problem is that it is very difficult to low-pass filter spatial data with current technology. But when we have 200+ MP sensors, this will be easy.

In the audio world, fully working oversampling has been available since late 1980's. The point is that you digitize at a really high frequency (e.g. the audio chips my company designs sample at 6.144 MHz), then low-pass filter at e.g. 20 kHz, then resample audio to e.g. 44.1 kHz. In this case, all frequencies up to 20 kHz can be represented PERFECTLY (as long as both depth is high enough). It might not look like it when you look at the waveform of a 20 kHz sine signal sampled at 44.1 kHz in an audio editing program, but when it is resampled and low-pass filtered in playback, it will come out as a PERFECT sine wave.

Link | Posted on Mar 29, 2017 at 16:53 UTC
In reply to:

lighthunter80: I'd rather buy a 5Ds for that kind of cash and adapt the 17mm TSE natively...

noirdesir: There are several 14 mm options available for the FF EF mount, and at f/2.8 or faster to begin with. So there are similar (and better) offers available for the 5Ds. Plus, the EF 11-24mm f/4L USM, a lens that doesn't have an equivalent in any other format or mount.

Link | Posted on Mar 24, 2017 at 22:51 UTC

Very interesting, Rishi!

Apart from a few minor typos ("a f/.." instead of "an f/.." in a couple of places), I think there is one place where there is a factual (math) error. Here:

"One thing does make us hopeful - recent conversations with some forum members alerted us to the fact that certain full-frame lenses, like the Zeiss Otus primes, actually project an image circle large enough for at least a square crop on Fujifilm's new MF format. That would essentially get you high quality F1.1 equivalent glass on the GFX 50S."

Not quite so. If you use a square crop, then the effective sensor size is 33mm x 33mm, which is only slightly larger than FF. The (reverse) crop factor is now 0.93, not 0.79, and e.g. an 85/1.4 lens installed this way would be equivalent to 80/1.3 on FF - which is hardly a relevant difference.

Apart from this one thing, I thought this was a really well thought of article of an interesting technical subject. Thanks!

Link | Posted on Mar 22, 2017 at 10:23 UTC as 215th comment | 2 replies

The little fanboy inside me just awoke and won't let me go on with my day before I say: "This is SOO cool!"

Thank you so much for sharing this with us, Canon and DPreview! Now I know where my L's came from, and I'm that much happier for it.

Link | Posted on Mar 21, 2017 at 07:45 UTC as 10th comment
On article Throwback Thursday: Canon EOS D30 (136 comments in total)

"5. I actually saw that happen once, but I was never able to confirm the story of another BDG returning his gazpacho soup because it was cold."

To see the gory details of this story documented in full video and audio look no further than the sixth and last episode of the first season of (the original UK) Red Dwarf. The episode's called "Me²", and it's one of the funniest and thought-provoking Sci-Fi episodes of all time! (Of course it is.)

And, yes, it's obviously Rimmer, Smeghead, who makes the demand to warm up his cold gazpacho soup. (Bonus: the ship's captain is the same guy that was leading LV426 in Aliens. (So he always dies after the incompetence of others? (Any more unnecessary references? (No!))))

Link | Posted on Mar 16, 2017 at 20:03 UTC as 10th comment
On article Fujifilm GFX 50S Review: Modern MF (899 comments in total)
In reply to:

villagranvicent: Too much DPR love for this camera. I wonder if this has anything to do with the fact that Amazon has it already on stock while the Hasselblad X1D has to be bought somewhere else.

Richard, serious question:

Have you considered taking the reference images in such a way that the FoV would be the same instead of image height? The current system gives a disadvantage to wider sensor cameras (in this case Canon and Sony). Personally, this has bugged me ever since the first Four Thirds cameras came out, i.e. for quite some time.

Link | Posted on Mar 14, 2017 at 20:46 UTC
On article Fujifilm GFX 50S Review: Modern MF (899 comments in total)
In reply to:

Peiasdf: Why are the two MF more zoomed in than the equally 50mp Canon and the just slightly smaller 42mp SONY image?

The three 50mp images should have the same FOV which is clearly not the case and the 42mp should only be 9% smaller linearly but appears to be 20% smaller linearly.

Frankly the test wasn't done properly.

The reason is that they keep the vertical framing the same, which favours cameras with the less wide 4:3 aspect ratio (these MF and e.g. m43 cameras) as opposed to cameras with 3:2 aspect ratio (e.g. Canon, Nikon, Sony DSLRs). If my calculations are correct, the 4:3 cameras are given a 12% area (6% linear) advantage this way.

As you mentioned, to make this test more fair the test procedure should be changed so that the shots are taked to keep the same FoV. That way objects in e.g. 50 MP images would stay the same size whether the camera has an aspect ratio of 16:9, 3:2, 4:3, or 1:1.

Link | Posted on Mar 14, 2017 at 07:02 UTC
In reply to:

Irakly Shanidze: How loud is the fan?

Plus, ">95 CRI" is really not much. It just means that there might be an indeterminate amount of the three primary colours, R, G and B. But absolutely no guarantees that skin tones or anything would look correct.

At home I have daylight ">98 CRI" tubes that cost a fortune, and they are ok. Not nearly perfect, but ok.

Link | Posted on Mar 9, 2017 at 22:09 UTC
Total: 336, showing: 1 – 20
« First‹ Previous12345Next ›Last »