Panasonic GH4 H.264 Codec Vs. Canon DSLR H.264 Codec - (Yes Canon is embarrassing)

[Guest Ebrahim Saadawi]

I’ve been reading quite a lot about codecs, compression and how different cameras compress the images coming out of the sensor to output the final video file. I’ve used Canon DSLRs for quite some time and never was disappointed. I trust them really. That was until my GH4 arrived lately.

After playing with the GH4’s 200 megabits/s and 50 megabits/s modes and finding that they are identical to my eyes, I kept wondering about how much these numbers really mean when it comes to evaluating codecs, and whether they're worth considering Especially since I’ve been seeing
24 megabits/s 4:2:0 8bit cameras give a comparable image to 800 megabits/s 4:2:2 cameras.

So I thought I would test the codecs I have in my Canon 70D DSLR and GH4, to see if they’re similar since they’re
the same codec, H.264, same compression rate, and same 4:2:0 and 8 bit, so the extent of compression, based on numbers, should be the same, right?

-Please note: I am not comparing the image quality between the Canon and GH4, I am comparing how they both compress their images.

-The way I though I would test this, is by taking a full resolution still image and compare that to the video file both cameras output under the same conditions. This should give a good idea on how the sensor output looks (Still) and how the final compressed video looks, thus give a fairly accurate representation of the Codec. Not a scientific test, full of flaws, but gives an idea of what the codec is doing. 

-I am testing the 1080p codecs, not the 4K one.


1- The Canon H.264 Codec Vs. Full resolution Still - Low ISO (100)

'' target='_blank'>>

2- The Panasonic H.264 Codec vs. Full resolution Still - Low ISO (100)

'' target='_blank'>>

3- The Canon H.264 Codec vs. Full resolution Still - High ISO (3200)

'' target='_blank'>>

4- The Panasonic H.264 Codec vs Full resolution Still - High ISO (3200)

'' target='_blank'>>


___________________________________

1-The Gh4 H.264-compressed files look nearly identical to the original full still image.

2-The amount of destruction the Canon H.264 is doing to the image is simply atrocious!

It's interesting to me because they're both H.264, 4:2:0 and 8 Bit codecs and both around 50 mbits/s 

So why are the results that different in terms of compression? I came to the conclusion that the Codec efficiency, how it's implemented, is much much more important than the spec-sheet numbers. This is why the C100 is giving great H.264 4:2:0 8 bit images, and the same reason why the GH4 image is that incredible. The compression is simply brilliant, with hardly any loss of detail and noise. 

If any one has any idea why can the same codec be much better in one camera, and much worse in another, please share, It would be nice to know what is the reason behind that efficient compression.

Both cameras output from the sensor very comparable image quality and maybe even better on the Canon, it's just that in video, Canon compresses the hell out of that image, and Panasonic doesn't. It's all about the codec, not the sensor or other technology. It's eye opening!

*Didn't compare the 4K to save the Canon blushes. The Canon DSLRs are going on ebay, I just can't stand how bad the image is after seeing the GH4's 

[maxotics]

I wouldn't sell your Canons quickly if you do a fair amount of photography in natural light, especially indoors.  

 

It doesn't surprise me that the JPGs from the GH4 and similar to the screen shots of the H.264 if there is no motion.  Most video compression quality is based on how well it stays to a low bit-rate when every pixel is changing in every frame (motion).

 

The C100 is built as a video camera so my guess is Canon has special chips and software that bin the pixels for optimum quality.  In other words, a C100 delivers better in-camera images to compression, over a comperable 5dII lets say.  Anyway, can we absolutely be sure compression is the difference between the C100 and consumer cameras?  A lot can go on between the sensor and final video compression.

 

If I could wave my magic wand, I would fix the number one misconception of many people on this forum--that there is only good compression and bad compression.  Like everything else, trade-offs are always chosen.  What's a good decision by Panasonic may be good for one situation, but bad for another.  Same for Canon.  If you're running things at Canon or Panasonic, do you pick high dynamic range over color depth, or color depth over high dynamic range?  Do you favor motion, or stillness?  Yes, I bet the GH4 is sharp as hell, but put a color rich subject in front of you Canon and see which color you like better.  

 

It's not about picking the best camera, the GH4 is not better than the Canons or visa-vera.  It's about picking the best trade-offs.  When I look at a study I want to see both arguments.  Even a 5DIII against an iPhone will lose in the "what is easiest to take a street photo with".  

 

Again, I'm not saying you shouldn't sell your Canons, only that you should know, and accept, the trade-off you'll make. 

[Nikkor]

As maxotics says, the problem is not the codec but the line skipping on the 70D. Anyway, Canon is a thing of the past unless they come up with something so superior videowise that it compensates all the benefits you get from a mirrorless camera.

[Vesku]

There are different levels of codec profiles in same 4:2:0 8bit h.264. The most recent and best profiles need a lot of real time processor power and gives better IQ at same data rate than "easier profiles". The real differences shows when there are lot of sharp movement in scene. GH3 50mbs codec is very good (almost like a poor mans RAW) and GH4 has even more complex profile and double the data rate.

 

Canon rebels has less horsepower than 4-core GH4 and Canon has less efficient codec. Canon also makes fullhd from 18Mpix sensor with less sophisticated scaler which causes aliasing, moire and softness. I think that Canon must throw away very much color information too.

[jcs]

The Canon C100 (and Sony FS700) 24Mbit/s H.264 codecs are on par with the GH4. It's not clear if the 50Mbit/s GH4 1080p can provide better quality vs. the 24Mbit/s codecs, however the image quality won't be as good as the other two cameras can produce higher quality at 1080p than the GH4 (GH4 needs to shoot 4K and downscale in post to match and possibly exceed those cameras (but won't be by much: the C100 is near the Nyquist limit for 1080p, the FS700 just a bit less). In my tests to far, it looks like the 24Mbit/s FS700 codec is doing a better job with color (and certainly dynamic range (14 stops for FS700 vs. 10-11.x for GH4)) than the 100Mbit/s GH4 4K (which provides more luma-edge resolution).

 

The Canon DSLRs are limited by low resolution images coming into the compressor (5D3), and aliasing (most of the other Canon DSLRs). Coupled with low CPU/ASIC power to perform high quality compression with H.264 and not-as-sophisticated-noise-reduction (as with the GH4), the final result from Canon DSLRs is comparatively low compared to cameras with better pre-compression frames and more powerful+modern compression and NR.

 

14-bit 11.x DR 5D3 RAW slays all these camera for final image quality, especially skin tones, at the cost of huge disk space and lots of extra work. If Canon were to release a new camera with modern CPU+ASICs and C100 level H.264 or better, they'd easily be competitive again. Apparently Sony is going to give it a go with the A7S (only issue noted so far is RS, which probably won't be any worse than the 5D3).

[Vesku]

The Canon C100 (and Sony FS700) 24Mbit/s H.264 codecs are on par with the GH4. It's not clear if the 50Mbit/s GH4 1080p can provide better quality vs. the 24Mbit/s codecs,

 

 

When shooting 1080/60P 24mbs is not enough to hold detail in complex scene. I think that c100 and fs700 codecs breaks also in very sharp complex motion in 24P. Real life like video quality needs 60P and GH3/GH4 are almost only cameras which has proper codec for that.

 

Try to shoot with 1/1000s with sharp frame and pan fast and see what happens with 24mbs codecs. 

 

GH3/GH4 1080P quality issues depends on scaler and image prosessing before codec. The codec is not limiting nothing but chroma resolution and of course bit depth.

 

Another matter what I am wondering is that I think GH4 line skips something because it introduces moire and aliasing and resolution loss in 1080P. Certainly in 96fps.

[Quirky]

Anyway, Canon is a thing of the past unless they come up with something so superior videowise that it compensates all the benefits you get from a mirrorless camera.

 

Video-wise, Canon already have something, namely their Cx00 line. Which, if not 'superior,' is decent enough, anyway, and 'mirrorless' enough. Albeit rather pricey, they're not quite a thing of the past yet. Today's GH4 hype is not likely to change that. All Canon have to do is to carry on updating and improving those, perhaps with another new model. Maybe a seriously re-designed EOS-M line for the mainstream market would be a nice addition. 

 

Their dSLR's are another story. As long as those are concerned, yes, I'd agree, there is not much point in investing in those any longer. 

[Andrew Reid]

The problem is Canon are not a video company at their core, they don't have the codec expertise or the number of video technicians as Panasonic have. Canon are happy to sell mass market stills cameras, but I worry they don't have the future insights due to their elderly top-brass management to stay with it long term, because the market is already shifting under them and they don't have an answer in a number of different areas. Compacts, consumer DSLR video, APS-C sensor performance and mirrorless being the main areas they have dropped the ball in recent years.

[hmcindie]

The codec has nothing to do with this. The softness is because of the quick debayering + line skipping. Even if you would put uncompressed out, it would be still as soft.

[jcs]

The codec has nothing to do with this. The softness is because of the quick debayering + line skipping. Even if you would put uncompressed out, it would be still as soft.

Right- Canon skips debayering altogether on the C100/C300 by averaging two green Bayer grids and taking red and blue as is from a 4K Bayer array.

 

That said, the 1920x1080 RAW Bayer array debayered with simple+fast bilinear (mlrawviewer) still looks much better than what comes out after Canon's H.264. It looks like perhaps they are adding a Gaussian/box blur before the compressor stage. The good news is that 5D3 H.264 sharpens reasonably well in post, though the color isn't of course as good as 14-bit RAW, especially if procesed with ACR.

[jgharding]

70D looks this way due to line skipping and poor scaling, not codec bits.

 

As mentioned, the C100/300 use low bitrate codecs but have incredible detail, as it's not a traditional bayer method. You get real 1080. No debayering, line skipping or poor scaling

 

Whack an external recorder on that bad boy and you're snapping at the heels of cinema cameras

[fuzzynormal]

Barreying, binning, codec, etc.

"Cut to the chase":

Which camera is best for walking around a major metropolitan area, grabbing random images, adding copyrighted music to the shots for an upload to Vimeo, and then having bragging rights that I've got the latest and greatest budget filmmaking gear?

GH4, right?

[Andrew Reid]

Stay constructive Fuzzynormal. That's clear flame bate and adds nothing to the topic at all.

[jgharding]

Of course, pan quickly in focus with a load of detail and bitrate really starts to matter more.

 

There are loads of "internal AVCHD vs Ninja ProRes HQ" tests for Canon C100 online that are useless, cos they're just still shots of relatively static subjects. In that situation, AVCHD will be fine, even with some grading, especially such a good implementation.

 

Under stress, bitrate becomes more and more important: so quick movement, moving detail, noise (also a form of moving detail)

 

Since bitrate is a measure over time, movement stresses it and is directly linked to it, as is complexity and detail in movement but not complexity and detail that's still so much.

 

Still detail is more affected by how the image is captured, scaled, debayered (or otherwise) and chroma sub-sampled (4:2:0, 4:2:2 etc). The exception being high ISO, where the noise (moving detail) affects resolution dramatically with lower bitrates. All the bits run out trying to keep track of the noise, and you run out of em.

 

For this reason, turning off noise reduction completely with compressed footage in very low light can sometimes be detrimental to the final image. YMMV, proceed with caution etc... On the C100, I sometimes use about 2 on the NR scale when recording internally. It gives the poor low bitrate a break!

 

Back to sub-sampling, do the C100 AVCHD vs Ninja test with a load of red patterned cloth and you'll have a mess in internal AVCHD, less messy in the Ninja, but never full 1080, because that's just what sub-sampling does. It's a bitch. Do it in low light with high ISO (about 10,000 or more for that camera) and you'll see something pretty horrible in the first shot. A sort of red low-resolution dancing mush.

 

That explains what AVCHD compression can do to (IMHO) the best HD feed available in a low-price camcorder. The bit-rate and sub-sampling together conspire to give your image a kicking in certain situations.

 

Your bit depth is your number of different colour shades. So more bits equals a more natural and rich result possible from the off. You can do a lot in the grade though with 8-bit, if you know your stuff, but more bits is nicer.

 

Arguably bit depth is the least important, simply because we're pretty forgiving of it in our visual systems without realising, and because all good post software pushes data around in a floating-point 32-bit space, where interim shades are generated to fill the gaps.

 

That said, who would turn down a higher bit depth source? You'd have to be bananas.

[Guest Ebrahim Saadawi]

Thank you for all the contribution. What I gathered is that I will not look to much into codecs next time I am evaluating specs, because as the C100/GH4/fs100/700 show us, H.264 4:2:0 8bit codec is very very good, that's also backed up by the fast that recording with a different (prores) codec with an external recorder doesn't improve much, so the internal h.264 is very good. What we should look into is the downscaling method (line skipping, binning) and debayering, sensor, color science, etc

Which camera is best for walking around a major metropolitan area, grabbing random images, adding copyrighted music to the shots for an upload to Vimeo, and then having bragging rights that I've got the latest and greatest budget filmmaking gear?

GH4, right?

Yes 100% right. For travel blogs/ travel videos and TV shows about cities for example the GH4 will work very well. as filming a city requires resolving high detail, with all the building textures, bricks, tree and leaves, colors and clouds, etc...

while with the Canon filming a city would give you aliasing/moire hell, plus the image is very soft for that kind of shots. Filming a high resolution scene with a Canon will look better if done with a modern phone. (something like Samsung S5 or Iphone)

-I don't get the music part though, I think it's irrelevant as both cameras will produce footage capable of adding music to. copyrighted or not.

_______________________

-I have another point too, what I found is that I don't always need the high resolution the Gh4 provides at all times.

I have a funny story: I remember the first time I used the Canon C line, the project involved lots of female faces shooting (make up stuff), I was incredibly excited by the form factor, sharpness in this little package. I happily shot the project with the camera, and my assistants shot with me using T2i T3is. The clients actually prefered the T2i image and said they look much "better", what ever that means to them! :D

I found that filming people's faces looks better with a lower resolution camera, and color is more important here too, which I found Canon to be superior in skin tones.

So when people their face on screen, they always prefer the soft, flattering look to higher resolution, accurate ones.

You can soften the sharp image in post though, so I will not say the Canon is "better" for filming people than the GH4, but at least, they're both good (after adding a decent bit of blur to gh4)

So of most of your work involves the filming people you're not losing much with a Canon really, in fact the slightly lower resolution image coupled with excellent skin color.production, will make your female clients very happy, more than the GH4 (well at least, if they are not filmmakers themselves ;) )

I believe it's the reason some people prefer the 2k Alexa over a 5k Red and say the image is more flattering and creamy.

[jgharding]

Yes your resolution comment is borne out many times over.

 

I actually use 1.0 of fast blur on portraits with C100 and C300 footage often as it's so sharp!

 

For landscape, I'll actually sharpen it.

 

Remember though, the external recordfer can help. I have a C300/Pix 240i project here with green screen. I'm glad we got ProRes.

[tugela]

Keep in mind that Canon still cameras use DIGIC processors, but the video cameras use DIGIC DV processors. They are not the same.

 

I believe the hardware encoder in both processors is the same however, but the difference is in how the processors read and process the output from the sensors. The DIGIC processors are optimized for stills, with video capability as an extra, but they are not particularly good at that. It is a compromise, with video being compromised in favor of stills. The DV processors on the other hand are optimized for video. The image that is generated by the processor goes through the same H.264 encoder, so if the C100 provides high quality images while something like the 5D does not, the difference is in how the processor deals with the raw data, not Canon's implementation of the H.264 encoder.