a few tech questions...

[Sam Eilertsen]

First of all, what actually determines the dynamic range of a camera? The dynamic range is, as I understand it, the maximum difference between brightnesses the camera can record, so if the shadows are at 0 IRE the highlights will be just out of clip. What actually affects this though? Sensor size? Color depth? Codec?

Second, is there anything besides sensor size (and exposure) that affects the low light performance of a camera? A lot of promsumer cameras advertise their "low light mode" or "superior low light performance", but I am suspicious that this is just a gimmick involving gain and noise reduction.

Finally, I've heard it mentioned that the "actual resolution" of a camera is less than the advertised resolution. How does this work? I assume it has something to do with the way Bayer patterns are merged into 4:2:2 or 4:2:0 color space.

And can someone explain what this graph, supposedly showing that DSLR's have <1080p resolution, actually means:

 

[Graeme Nattress]

Dynamic range is determined by many factors, starting with the sensor, it's design, analogue and digital circuitry, any image processing that follows the sensor. Compression can effect dynamic range, so that has to be carefully designed to avoid compromising the dynamic range of the image too.

Low light performance is a combination of the sensitivity of the camera, and everything that also effects dynamic range. Indeed, because low light performance often hangs on by how much you can grade-up or brighten an image, it's very dynamic range dependent.

Low light can also be influenced by infra-red response and indeed some consumer cameras let through IR to boost low light performance at the expense of colorimetry, or even have a IR illuminator to allow them to literally "see in the dark" albeit in black and white.

Actual resolution is what is measured and tests the resolution of the full camera system, from lens through the sensor, image processing, compression to the final image.

Due to the necessities of how sampling theory works, if you attempt to sample (the sensor is doing sampling of the image) a higher resolution than the resolution of the sensor, you will get an unwanted artifact called aliasing, and aliasing leads to visible moire and other image artifacts. It is therefore important to limit the resolution of the scene reaching the sensor so as to avoid these artifacts or minimize their visibility in practical use.

The graph is to show how much image contrast the system sees at all resolutions (or levels of detail) up to 4k. Amplitude of the signal is on the vertical axis and resolution on the horizontal axis.

The graph for the DSLR starts strong at the low details, as you'd expect, but it's modulations stop around 1.5k which is indeed sub 1080p - which is 1920 on the horizontal around 1.9k. The interesting thing is that you'd think that once the resolution runs out, you'd just get a flat-line, but you can see around the 3k mark resolution appears again. This is not real resolution but the aliases mentioned earlier where high frequency detail "folds back" to appear in the image as low frequency artifacts as moire.

The red graph in the background shows a RED camera pointed at the same chart, and as you can see there is very strong detail out to 3k, and then a gradual tailing off to very little at 4k, which keeps any aliasing or moire at a minimum leading to a low artifact image, compared to the high artifact image you'd see on the DSLR.

Graeme

 

[Alexander Ibrahim]

The truest answer I can give you is that your questions require long study of photography and cinematography. I can give you a survey of the answers, but really you need to do a lot of studying.

In other words, its beyond the scope of a forum like this... but people do try. There are a lot of conditional statements and simplifications in what I am about to write. It may take a few reads to digest it all- and some of it isn't strictly true.

It should definitely be enough for you to start learning more on your own.

Yes, most DSLR camera's don't really deliver 1080p resolution.

Most cameras record a format. For example 1080p or 4K. That is just how many pixels it will record information for. What the camera actually sees is a different matter.

The chart you refer to is trying to show the contrast level of a resolution chart.

Resolution charts are black and white (theoretically). As the lines get closer together and start to exceed the resolution of the camera they will represent the white and black resolution chart with gray tones. When the plot levels out on the chart what you are seeing is all balck and white detail past that ersolution is being rendered as gray.

This is a limit of the optical system of the camera together with the sensor resolution.

On DSLR cameras another limiting factor is that the entire sensor is not read to create images. Current DSLR cameras can't read the whole sensor 24 times a second... so they have to "cheat" by skipping parts of the sensor. In fact for DSLR cameras this line skipping is the dominant factor in their reduced resolution.

Done intelligently this sort of line skipping doesn't have to be the huge problem that it is on current DSLR cameras. For example, when you shoot a RED One in 2K mode, RED skips reading most of the sensor... its just that they choose to read the part of the sensor they are using as a continuous block.

Ostensibly, if they could read the sensor fast enough, most DSLR's would be able to produce high quality 4K or higher resolution images... they already do this for stills. Assuming the manufacturers optimize for sensor read out speed, without trying to improve other characteristic of the sensor, there could be a commodity priced APS-C full frame 60 fps sensor in the next two years.

If this was the case then, like RED and film cameras, lenses and filter quality would be the limiting factors on camera resolution.

In any case, I find my 7D delivers higher than 720p in nearly all circumstances, and usually higher than HDV (1440x960 or 1440x1080) resolution. This is in rough concordance with the data shown on the chart.

If you look at the chart, you'll see that the RED delivers about 3.5k of effective resolution.

Low light performance of a camera is dominated by lens aperture and sensor sensitivity. The T stop of a lens pretty much describes the lens side of the problem. So let me focus on sensor issues.

A sensor has two characteristics of major concern to light sensitivity.

First, and most fundamental, is the innate light sensitivity of the technology in play. How much of the light that hits a sensor can be converted into image data.

Second is the sensor's size. More accurately how big are the individual photosites of a given sensor. A photosite can be thought of as a single pixel. There are tricks that manufacturers use to make the photosites of the sensor function as if they were bigger than they are, for example a microlens placed over each photosite and so on.

As a potential user the details of most camera's sensor technology are often difficult to discover. That's because manufacturer's keep them secret- because their secrets are what give them a competitive advantage.

So what do you do? For most video cameras I usually establish an ISO equivalent for the various settings the camera offers. For example an HPX-170 set for 720/24p using the "Cinelike-V" look and 180 degree shutter is roughly equivalent to ISO 225. This is done by testing with a light meter and waveform monitor.

Dynamic range of a camera is the difference in lighting level it can see at once. So, for example a camera might see 3 stops below key into the shadows, and 2 stops above key into the highlights. That camera has a dynamic range of 5 stops.

A RED One typically is evaluated to deliver about 11 stops of dynamic range, and thanks to FLUT you can set where you would like the midpoint to be. A Sony EX1 has about 9 stops. A Panasonic HPX-170 has about 7 stops. A Canon 7D has about 10 stops.

Do your own testing under the lighting conditions and set up you plan on using- your evaluation may be quite different than mine or anyone elses.

The best place to start further research on this topic:

Panavision's Demystifying Digital Camera Specifications

 

[Sam Eilertsen]

Thanks a lot for your responses, they are really helpful.

One other question: I have heard the standard definition DV camcorders perform better under low light than HDV, simply because more light hits each pixel. Will 4K cameras suffer a similar disadvantage in low light, compared to HD cameras?

 

[Graeme Nattress]

There's no absolute rule - you've got to test. Pixel size is a factor, but so is pixel design and accompanying electronics.

Graeme

 

[David Mullen ASC]

Also remember that 4K cameras tend to have 35mm-sized sensors while HD cameras tend to have 2/3" sensors, so the actual difference in photosite size may not be that much... to oversimplify, imagine you are putting twice as many photosites in a row but that row is twice as wide as before, so the size of each photosite can be the same.