What factors should be considered when choosing a digital camera?
With any camera there are pros and cons. What you need your camera for will decide which order you will consider the following qualifiers. For example; (these terms will be discussed throughout the document)
-a documentary camera would benefit from inbuilt ND filters, small bitrate, and good audio components
-an action cinema camera would be better from large file sizes and low rolling shutter
-any shot needing a wide angle of view would benefit from a larger sensor size
-an event camera would greatly benefit from low file sizes, duel memory card slots, a smaller sensor, and higher light sensitivity
This article is not in an order of importance but more of an order of complexity vs importance.
CROP (1)
The first thing to know is your camera’s sensor size. Most DSLR cameras are APS-C sized (similar to super35), about 1.5x-1.6x crop factor which means it’s about 62.5% of the size of a full frame sensor. There are also micro 4/3 sensors in cameras like the Panasonic LUMIX GH* range that are smaller once again with a crop factor of 2x (50%). The black magic camera has a crop of 2.39x and the pocket has 2.88x. The crop number is representative of how many times you need to multiply the focal length, ISO, and the F-stop of the lens.
Lots of people argue over the usefulness crop size, but here’s the pros and cons;
– If not designed right, a smaller sensor will have poor low light capabilities and dynamic
range (though cheap larger sensors can be worse, so a rule of thumb is $=quality)
– It’s harder to get a wide angle of view, and not going wider makes it very shaky even with in camera/lens image stabilisation
– The smaller the sensor, the harder it is to get a shallow depth of field because of focal length restraints and F-stop multiplication. This can be a pro, because lots more will stay in focus in the image (particularly advantages for documentary work)
– They are cheaper to make (this does not mean the camera has to be cheaper, the GH*
cameras are more expensive because they have more options standard from factory)
– A smaller sensor is likely to have less rolling shutter (jello movement) (this is a general rule, but not always correct)
Figure 1
(taken from a black magic pocket cinema camera and canon 50D by Phillip Cook. Top:Canon 50D 28mm, middle:BMPCC 16mm, bottom:the two images overlaying)
These photos are taken from a camera with 2.88x crop at 16mm focal length equalling an equivalent 46mm (2.88×16=46) field of view on a full frame, and the other photo is 1.6x crop at 29mm giving the same 46mm full frame field of view.
Having a larger aperture gives you a shallower depth of field, and aperture is the focal length divided by the diameter of the iris. So as you crop, the focal length changes, meaning the f-stop number grows, giving you a larger depth of field. To calculate this, all that has to be done is times the f-stop by the crop factor as well to know what the equivalent depth of field is.
For further explanation, follow this link to an informative video, http://youtu.be/DtDotqLx6nA?t=14m59s
Crop can be over come on certain cameras like some Sony, GH* cameras, and the black magic cameras with a MFT mount, by using a speed booster. There are advantages to speed boosters too, not only do they make the image wider, there for meaning you can use a more zoomed focal length, but it will also give you extra light from your scene. Many recent cameras also have the ability to film in crop modes, and this also usually gives less rolling shutter but the same light sensitivity.
LOW LIGHT SENSITIVITY AND DYNAMIC RANGE (DR)
The common thought pattern is that a larger sensor will give you better light pick up per photosite and dynamic range. But it also is effected by the types of materials used to make the sensor itself, the gaps between the individual photosites, the in camera noise reduction, and how many megapixles the sensor is divided into.
An advantages to good low light sensitivity is the ability to stop down your lens. Because a lens at its widest open aperture is always softer than a few stops closed from it’s widest setting (aperture is measured in stops, and stopping down means making it smaller but confusingly gives a bigger number). Stopping down a lens gives you a larger DOF and less light. With a larger sensor however, you can keep a smaller DOF because you can use a longer focal length. ( NB: stopping down too far will also make the focus soft)
Other advantages includes using higher frames per second which divides up the light in any given second into smaller amounts, making it darker per frame.
The final advantage is the most obvious, the ability to shoot in darker settings with less lighting requirements. At the time of writing this, the Sony A7s and the Nikon D4s are the best low light DSLR cameras and can even use less light than moonlight. The majority of actual cine cameras are also quite good, a few good examples are; (starting at the cheapest) Blackmagic Cinema Camera, Canon C100, Sony FS7, Sony FS700, RED SCARLET-X M-X, and the Canon C300 are a few good examples.
Dynamic range is the ability to get as much darkness and light in the same frame before it gets “clipped” (information that is stored as exactly black or exactly white with no colour information). DR is measured in stops, that is when a lens is stopped so something is full black and how far you stop it up until the same part is full white.
Having a high dynamic range is good for filming inside and having a window in shot and not have the outside blown out (information stored as full white), or filming from a well lit area and needing to see something in a dark corner without ISO noise. (I say ISO noise because if something isn’t black, you can “raise” it in post production, but the darker it is, the more visible ISO/dB noise can be seen)
When basing a decision on a camera’s DR, it is also important to check its native ISO, which is the ISO level at which that camera’s maximum DR has been measured. This can be problematic if the native ISO is too high or too low and should be considered and may require finding internet videos that are in similar circumstances to what you are shooting, or actively testing the camera in person.
Another thing that should be taken into consideration is what kind of compression it uses, because raw will give not only better low light performance but also a higher DR. Below is 2 images that were taken at same time but the top was taken as a JPEG and the bottom was shot as RAW and both were adjusted to show their comparative DR:
Figure 2
(taken by Phillip Cook on a Canon 50D, top:jpeg, bottom:RAW)
They Both have the same curves adjustment, but you can see more detail in the trees and the darker objects with RAW (compression and RAW will be explained later).
A higher dynamic range is mostly achieved by a specific “log” type colour profile in the camera, which gives a flat look that has to be stretched back out in colour correction and this is usually a good thing, but if the dynamic range is too large and recording bitdepth is only 8bit (explaned later), then it can be ruined in post.
ROLLING SUTTER & SENSOR TYPE
Rolling shutter is the result of how an image sensor’s information is processed. This was not a problem with CCD sensors because their image was sent from the sensor in one continuous analogue stream to the processor and precessed at all once. But a CMOS or similar sensor will have an individual digitising processor attached for each pixel (2), and is then sent to the processor to put the image together. With a CMOS sensor, the image is sent one horizontal line at a time digitally, starting at the top.
The disadvantage of the CCD style sensor is an analogue signal is prone to interference, where as the CMOS sensor almost entirely eliminates the path that the analogue signal has to take. This gives the CMOS sensor a much cleaner (less noise) image. Rolling shutter can be measured in milliseconds, from the time in which the top line is taken to the bottom line (3).
A CMOS sensor is considerably cheaper to make, uses less electricity, and generates less heat than a CCD sensor.
Some CMOS cameras have a “global shutter” (ie. Blackmagic 4k and the Sony F55) where it takes the image in one go similar to how a CCD would. (12)
The side effects on rolling shutter is mainly seen when panning, tracking, and objects moving fast across the front of the camera, these things will show a lean in an image. The other instance where you will see it is during hand held, because when the camera makes a vertical movement it squashes when moving down and stretches when moving up. (Examples for almost all cameras can be found on youtube)
Below is an example of a moving train with rolling shutter in contrast to a non-moving person.
Figure 3
(from https://vimeo.com/103951451 uploaded by John Sobek, shot on a Sony A7s)
In figure 3 it is also worth noting that because of paralax and things that are closer the camera move faster in the image, making the front of the train more slanted than the inside of it.
COMRESSION
The next and also quite important thing you have to look at is the type of compression and how much it compresses. The easiest thing to do is to judge the quality of a cameras codec is to look at the bit rate (not to be confused with bitdepth). Sony have nice cameras like their A, VG, and FS lines, but their bit rate is about 24 megabits (not megabytes, 8Mbit=1megabyte) per second. The GH4 has from 17 to 200. Canon DSLRs start at about 45mbps and goes to over 400mbps, and Nikon (which otherwise are great cameras) have a standard bitrate of 24mbps. Both Nikon and Canon have “firmware hacks”, but the canon hack is much larger, and has the option to shoot raw video (this will be explained later).
The general way to compress is to use block-oriented, motion-compensation-based compression, and discard anything deemed too bright or too dark (and example is figure 2).
The motion-compensation-based (known as Long-GOP) part of compression is more widely used right now, and is better if you are concerned with colour grading and smaller file sizes, but worse if you are doing digital effects (including green screen). It simply (though uses either complicated algorithms based on changes in frames and maintaining a bitrate or simple timed keyframes) takes two frames called “I” frames, and the frames between are called either “P” or “B” frames (depending on what’s happening in the picture). More information (5) & (6).
Block-oriented compression is in reference to the Long-GOP prediction method with chroma subsampeling that is explained later in. (16)
Using an all I frame codec, like in modern high-end Canon DSLR and cameras, PRORES, and DNxHD will give you a slight increase in sharpness too.
Spatial compression is a common type of compression in lower end codecs or delivery codecs, they reduce the file size dramatically, but also make image manipulation difficult (figure 4 & 5). It is the operation of dividing up an image into blocks where colours are similar enough to be considered the same.
Top: Figure 4, Bottom: Figure 5 (figure 4 is a standard compressed image, and 5 is that image with a hue shift that reveals the hidden spatial compression artifacts)
Codecs rely on you giving it some external input, ie white balance (not needed for RAW) and an ISO number (ISO is an analogue amplification and gives less noise than digital amplification).
There are other ways to compress, but these ways are used in H.264 and is the most commonly used type of codec at the moment and even used in other codec types (competing types are apple prores, XAVC, and DNxHD that works similarly). There are downsides to compression, you lose dynamic range, it’s a hindrance to colour grading and most of all reduces the actual resolution (meaning smaller raw resolution can have more detail than larger but compressed resolutions), but all has the upside of smaller file sizes (which can be very important). To not have compression you would have to shoot in raw video, there are many file types for raw, but all do basically the same thing, it stores every pixel of every frame, the downside is that you need a camera that is fast enough, and expensive memory cards. The 3 main forms of camera storage is SD (Secure Digital) cards, CF (compact flash) cards, Cfast cards, and SSD (solid state drive) like what is used in computers. Also with the faster data rates of raw, it generates lots of heat, so working in hot or confined places can cause problems, and uses an extremely large amount of space, where a normal compressed video may use 300 megabytes for a minute of 1080, raw would use about 5,220 megabytes in the same minute. The final downside of raw video is that you would need a computer that can deal with such large files at a fast rate, or editing will take a long time (there are cheats to reduce this though, which include using proxy files while editing). RAW also keeps information that is considered above white, and below black, giving a higher dynamic range and maintains colour information too.
It is worth checking any prospective camera for it’s external output types are. Some outputs give you a RAW video feed from a HDMI port. Other output ports include the industry standard SDI port which is used because of it’s ability to lock the chord in and solid metal connectors.
Apple keep prores a secret, but a similar codec called DNXHD works by taking a stream of JPEG images, mainly utilising algorithms to shorten binary code lengths giving an almost lossless compression but is limited (like prores) to 10bit and 8bit bitdepth. (7)
AUDIO
Most cameras come with some kind of audio input, and your input type is also important too. Very rarely do you find a camera with an internal mic worth worth using, and is usually only there for synchronising your footage with other sources (including automatic synchronisation software for multicam set ups or separate sound recording). Most documentary style cameras come with mounted microphones that are quite good, but professionals tend to use off camera microphones anyway.
Most audio internal recording formats are fine, but something to consider when deciding your audio solution is whether you can just plug a mic in (like a simple rode videomic) without a preamp. For instance, the BMPCC has very inadequate internal preamps, so you would want to decide what you would do there.
Most DSLRs have a 3.5mm jack, and headphones plug, the BMCC has 2x 6.5mm jacks (6.5mm is not a standard in film making and unusual), XLR plugs, and phantom powered XLR plugs for condenser microphones. Before choosing to use internal recording (with internal or external mics), you should always make sure that the camera you have has audio readout levels, because you could potentially get bad sound when not kept in check.
External recording is a very professional way of doing things and always requires a person in charge of it. This also gives the option to laden the camera with less cables. Having a person to do this takes a job that’s not important to the camera operator away from them. Having a skilled person doing this for you is always the best choice!
BITDEPTH
Bitdepth is how many bits (1s and 0s) are allocated to each colour or chroma value
(chroma=lightness value). The calculation is 2 to the power of the bitdepth value;
8bit = 2^8 =256 colours
The majority of cameras (DSLR and most cheaper cini-like cameras like the Sony FS and Canon C line) use 8bit, and the 256 values are then selected from a predetermined pallet of up to millions of colours. 8Bit is also the bitdepth that you would deliver to air on TV, but 10bit is also commonly accepted. There are different ways to implement a bitdepth, and Philip Bloom was quoted saying that the Canon C300 was the best implementation of 8bit that he had ever seen.
The downside of such a low number of colours to choose from is called banding, where in a smooth gradient, lines are created. (7)
10bit =2^10 =1024 colours
The second most used bitdepth because it is used in both prores and DNXHD they consider it an acceptable amount of colours. Many external recorders record 10bit, even if the input is only 8bit. The GH4 is the only current (at the time of writing this) DSLR to have this as an option, even if it is only through the micro HDMI port without an expensive add-on. All of the Blackmagic cameras offer 10bit internal recording to prores and DNXHD.
As stated on the top of page 11, “TECHNICAL STANDARDS FOR DELIVERY OF TELEVISION
PROGRAMMES TO BBC”, they accept 10bit and 8bit formats. (8)
12bit =2^12 =4096 colours
The Blackmagic range offers internal RAW recording at 12bit
14bit =2^14 =16384 colours
Using the MagicLarntern hack on a Canon DSLR to get RAW video will operate at 14bits. The majority opinion is that this is too high (17) for what most people need, but are unable to implement a hack to give a lower bitdepth in raw video (at the time of writing this).
16bit =2^16 =65536 colours
RAW usually operates at 12bit and above, but not most cheaper cameras, ie the FS100 and all enabled DSLRs output only 8bit RAW, even if the output is 4.4.4 or not.
8bit is the minimum that will be accepted by most companies (8), even if it’s black and white (that gives 256 gradients). However 8bit suffers from banding and makes colour grading difficult, it also makes selecting specific colours difficult if you are colour grading specific colours or chroma keying (blue/green screen). Though even with it’s limitations, it is the most used bitdepth because of it’s data size to quality ratio.
4.4.4 4.2.2 4.2.0 4.0.0 CHROMA SUBSAMPELING (13)
This is a way to express chroma subsampling compression/Bayer pattern sensor interpolation (because all current single sensor cameras on the market are Bayer pattern we will look at this in the context of compression). This takes advantage of the fact that the human eye is much more sensitive to light than it is to colour and when compressing it stores more light information and applies adjacent colour information to make a pixel with less MB of information.
The first number in the three number sequence is how many horizontal columns of pixels are being described (so not entirely important). The usual way compression in this sense works is to repeat the subsampling after every second line, so the second number explains how many of the initial number are pixels contain colour information and the third number explains the same thing.
So in something 4.2.0 (most common amongst DSLR and other cheaper cameras) there is 4 columns, and the top row of those 4, only 2 have colour, and the next row is only light information. Because we are more sensitive to light than colour, this works fine for viewing but can be problematic for post production work when trying to define an edge or change colour information. Though this does however help in reducing moire (like the colour noise in figure 6).
Figure 6
(taken by Phillip Cook with Devon Smith on a BMPCC, using RAW at ISO 1600)
These are easily explained as 4.4.4 results a larger file type usually found in RAW files giving a better picture. 4.2.2 is smaller but quality is debatable in relation to its file saving size if you are not doing any chroma keying work (blue/green screen), and is most commonly found with prores and DNXHD. 4.2.0 is the most commonly found type of chroma subsampeling and is adequate for even viewing on a small cinema screen. 4.0.0 is black and white.
4.2.0 is found in almost all DSLR cameras, but often is not the case when using an external output video feed to an external recorder.
Reference material
http://en.wikipedia.org/wiki/Chroma_subsampling
http://books.google.com.au/books?id=c-MjAwAAQBAJ&pg=PA29&dq=bayer+pattern&hl=en&sa=X&ei=guEBVJOxG428uATE6YLYBQ
&ved=0CBwQ6AEwAA#v=onepage&q=bayer%20pattern&f=false
DOWNSAMPLING/ALIASING AND MOIRE
Aliasing is caused by a resolution too low to pick finer detail or downsampling and can cause jagged lines on objects that are straight.
Figure 7
(a frame from Makhala’s Flinders University film “to Slay a Pig”)
In the above image you see lines that would otherwise be straight but because of the sensor’s downsampling “decimates” the 2 megapixles chosen from the origional 18 megapixles the sensor has. Selecting less than the full amount of pixles with this method causes jagged and uneven lines.
Figure 8
(this represents a diagonal light source on a sensor, and picking pixles every 4 rows down (blue) and the apparent non-straightness of it represented in pink)
Aliasing is also cuased by regular compression and subsampeling causing it to look like the pixels are larger.
Figure 9
(left is an example of low amounts of compression, and right is high compression)
Moire is caused by a similar issue of light shining on a sensor and not covering squares perfectly. Moire is caused by light not hitting all the photosites in a whole pixel. In the image bellow, the white light is only hitting the red and green of a typical Bayer pattern sensor, causing the sensor to think that the colour of that pixel is something other than white, and this will happen along the edges of objects.
Figure 10
(an example of how moire occurs)
This is not usually an issue in cameras with with a sensor for each colour, ie. 3CCD cameras
because each photosite is the size of the pixel and not sharing the space of a single pixel.
There are filters to reduce both moire and aliasing, either digitally using interpollation of the surrounding pixles (9), or more commonly with an optical lowpass filter (or blur filter) (10) (11). There are options to reduce aliasing and moire, and that comes down to how the sensor is designed, for example, making the total resolution of the sensor devisable by the used resolution of the video image equally vertically and horrizontally. (moire is often coinsidentally fixed by subsampeling)
When choosing a camera, things that should be considered is if the aliasing and moire in the image is too harsh for your specific purpose. Then wether it already has an anti-aliasing filter (also called an optical lowpass filter) already or if one can be added.
RESOLUTION
The current standard resolution amoungst cameras is 1920×1080 pixles making aproximately 2 megapixles, and cinema standards are 2k (2048×1080) and 4k (4096×2160) (14), plus UHD (3840×2160).
Australian television and other contries still how ever broadcast mainly or entirely in standard definition 1024×576 (15).
Even if you are not delivering at higher than HD quality, there is still advangates of higher
resolutions if you can afford the extra storage and editing times (UHD is the same as trying to watch 4 movies at the same time because there’s 4 times the pixles in the image). The main advantage is the ability to adjust your framing in post by cropping in on your image substantually and still maintaining HD quality.
OUTPUTS, ND FILTERS AND EXTRAS
When looking at a camera, most will offer better quality footage out of an output and if you are not satisfied with the internal recording, it’s worth looking at the specifications of the output. The other thing to check with output, is what types of output there are, micro HDMI is easily damaged, so is mini and normal HDMI, which is why many bigger/better cameras use an SDI connection. SDI can also use a longer cable.
ND filters (neutral density), makes things darker and can be inbuilt. Most people will eliminate many cameras from their list of choices based on whether they have internal ND filters or not (making filming quicker can be very important). Your alternative options for ND filters is to have a filter box, but usually requires the camera to be on rails, also there are plain screw in ND filters, or a screw in adjustable ND filter which is problematic if the end of your lens rotates when you focus or you are using a polarising filter (to remove glare from windows amongst other things). It is also important to note that not all cameras have the same amount of filters and how much they enervate the light in stops.
Focus peaking is a useful tool for gaining focus before and during shots, it illuminates parts of the image that are sharp. This can be used during filming too, removing guess work.
Figure 11
(an example of focus peaking on the Canon 50D with magiclantern 2.3)
Ergonomics can also be an important factor. Aside from weight which can be important if you have hand holding the camera for long periods of time, but you have to find out if you will need a rig to hold your camera and how much that will add onto the cost of the camera in question of buying or renting. Things that may need to be added include a matte box which requires rails, a top handle, shoulder rig, microphone, or an external battery if the supplied battery is not good enough.
JE NE SAIS QUOI
If a camera ticks all the boxes then it will give you a good picture, but there is a saying, “if it looks good, it is good”. Meaning that if you like the look it has, questioning it may sometimes just be a waste of time.
Parentheses
(1) http://en.wikipedia.org/wiki/Crop_factor
(2) “Real World Digital Photography” By Katrin Eismann, Sean Duggan, Tim Grey
(3) http://www.dvxuser.com/V6/showthread.php?303559-Measuring-rolling-shutter-put-anumber-
on-this-issue!
(4) http://en.wikipedia.org/wiki/Photoreceptor_cell
(5) http://en.wikipedia.org/wiki/Motion_compensation
(6) http://en.wikipedia.org/wiki/H.264/MPEG-4_AVC
(7) http://en.wikipedia.org/wiki/Colour_banding
(8) pg 11 “TECHNICAL STANDARDS FOR DELIVERY OF TELEVISION PROGRAMMES
TO BBC” http://dpp-assets.s3.amazonaws.com/wpcontent/
uploads/specs/bbc/TechnicalDeliveryStandardsBBC.pdf
(9) pg 494, “Photoshop CS3 Bible” by Laurie Ulrich Fuller, and Robert C. Fuller
(10) (2) pg 29, “Real World Image Sharpening with Adobe Photoshop, Camera Raw, and
Lightroom”, by Bruce Fraser, and Jeff Schewe
(11) http://en.wikipedia.org/wiki/Anti-aliasing_filter
(12) part 4.3.1 of “Smart CMOS Image Sensors and Applications” By Jun Ohta
(13) the truth about 4k by John Galt
https://library.creativecow.net/galt_john/John_Galt_2K_4K_Truth_About_Pixels/1
(14) pg 117, “High-Quality Visual Experience: Creation, Processing and Interactivity of
High-Resolution and High-dimensional Video Signals”, By Marta Mrak, Mislav Grgic,
Murat Kunt
(15) pg 2, “Free TV Australia Fact sheet” http://www.freetv.com.au/media/Free-to-
View_Digital_TV/DIGITAL_FACT_SHEET_April_2009.pdf
(16) pg 13, “The Handbook of MPEG Applications: Standards in Practice”, edited by
Marios C. Angelides, Harry Agius
(17) http://www.magiclantern.fm/forum/index.php?topic=5601.0
- This paper was written in 2014 for a research topic at Flinders University, South Australia. This was copied from the PDF I submitted, so there may be formatting issues.
Phillip Haven Cook 