DJI Mavic, Air and Mini Drones
Friendly, Helpful & Knowledgeable Community
Join Us Now

Video is worse quality than photo - Why is that?

Yaros

Well-Known Member
Joined
Aug 23, 2021
Messages
1,231
Reactions
1,182
Location
Sa Coma, Mallorca
So, I have noticed a while ago that the Mavic Air 2 takes incredible quality photos, even in the normal 12MP mode.
But when it comes to video, it's a bit different. I didn't think much about it, but now I'm wondering, why is that?
I took a video in 4K Wide 60FPS Normal Mode, D-Cinelike Color Profile, MP4 H.265.
Then took a photo in Normal Mode, 4:3 Aspect Ratio, then compared them. The photo is at least 1.5 times sharper.

Should I use MOV format for video instead?
How to get sharper video on the Mavic Air 2?
Why doesn't DJI introduce a sharpening option on the Air 2 as they did on the Mini 3 recently?
 
What were the shutter speeds for each? Motion blur may have caused the softening of the video images.
 
You're comparing apples to oranges, and wondering why they don't look or taste the same.

How are you comparing them? Have you extracted a still frame from your video and used that to compare? Or are you looking at a short video clip? How are you viewing? monitor resolution?

There are so many variables here, where to begin?
 
So, I have noticed a while ago that the Mavic Air 2 takes incredible quality photos, even in the normal 12MP mode.
But when it comes to video, it's a bit different. I didn't think much about it, but now I'm wondering, why is that?
I took a video in 4K Wide 60FPS Normal Mode, D-Cinelike Color Profile, MP4 H.265.
Then took a photo in Normal Mode, 4:3 Aspect Ratio, then compared them. The photo is at least 1.5 times sharper.

Should I use MOV format for video instead?
How to get sharper video on the Mavic Air 2?
Why doesn't DJI introduce a sharpening option on the Air 2 as they did on the Mini 3 recently?
Make sure you have a large screen that can handle the 4K quality and clean you lens from time to time
HTH
 
So, I have noticed a while ago that the Mavic Air 2 takes incredible quality photos, even in the normal 12MP mode.
But when it comes to video, it's a bit different. I didn't think much about it, but now I'm wondering, why is that?
I took a video in 4K Wide 60FPS Normal Mode, D-Cinelike Color Profile, MP4 H.265.
Then took a photo in Normal Mode, 4:3 Aspect Ratio, then compared them. The photo is at least 1.5 times sharper.

Should I use MOV format for video instead?
How to get sharper video on the Mavic Air 2?
Why doesn't DJI introduce a sharpening option on the Air 2 as they did on the Mini 3 recently?
Photos are 12MB images, whereas still frames from 4K video are at most 3MB. You'll have roughly twice the sharpness and resolution in still frames extracted from 4K at 30fps, because the maximum bitrate has to be distributed among the frames per second. Shutter speed is also relevant, if there is motion blur causing the lack of sharpness. Don't use ND filters. They deliberately create soft video still images.
 
  • Like
Reactions: Felix le Chat
So, I have noticed a while ago that the Mavic Air 2 takes incredible quality photos, even in the normal 12MP mode.
But when it comes to video, it's a bit different. I didn't think much about it, but now I'm wondering, why is that?
I took a video in 4K Wide 60FPS Normal Mode, D-Cinelike Color Profile, MP4 H.265.
Then took a photo in Normal Mode, 4:3 Aspect Ratio, then compared them. The photo is at least 1.5 times sharper.

Should I use MOV format for video instead?
How to get sharper video on the Mavic Air 2?
Why doesn't DJI introduce a sharpening option on the Air 2 as they did on the Mini 3 recently?

Photos and videos are different. You cant compare them. Video wants some motion blur or the footage will look like a soap opera.
 
There is zero difference between .MP4 and .MOV

But there are differences in several of the other settings.

On a mini three pro, the following settings:

.mp4 and .mov are no different, both are 4:2:0

h264 - high profile
h265 - main profile

30fps and lower: allows H264 high profile
Above 30 FPS: forces h265 main profile

Normal: 8bit, Rec709
DCinelike: 10bit, pseudo log

jpeg still: 8 bit
DNG still: 16bit
 
  • Like
Reactions: GadgetGuy
h264 - high profile
h265 - main profile

30fps and lower: allows H264 high profile
Above 30 FPS: forces h265 main profile
So 30FPS H.264 is better quality than 60FPS H.265?
 
So 30FPS H.264 is better quality than 60FPS H.265?

I am not going to say that specifically, as I haven't done enough tests.

However, there is a maximum data rate, and as the frame rate goes up the data rate limits how much data is allocated to each frame.

Considering data rate only
Just shooting a static test frame with no motion at 4K, 24 FPS gives a data rate of around 100mbit/s, increasing the frame rate to 48 FPS data rate goes up, but only to about 115mbit/s. Twice as many frames but only 15% more data. Now this is for shooting a test chart with no motion in the frame, motion frame will change the date rate significantly. The data rate on the Mini 3 Pro is limited to 150mbit/s.

If you are shooting at 60 FPS and your data rate is 150mbit/s (19 MByte/s), that is 2.5mbit per frame, and optimized for motion so you have key frames (I frames) and then frames that are made up of only the amount of change from the previous frame (P frames).

2.5 megabit per frame is 0.31 MegaByte

Taking a single JPEG is 3.8 MByte for that one frame, more than 10 times more data, and optimized for a still image. A single raw image is 24 MByte.

IMPLICATION: the jpeg has 16% of the original data, and the h265 frame has only 1.6% of the original data. 😳

Considering codecs
Modern image data compression is based on the perceptual qualities of the human vision system (HVS), and the HVS has these qualities:
  • Contrast, edges, and fine detail are perceived from the luminance (lightness/darkness) of a scene, in V1, the first stage of the visual cortex.
  • Color, meaning hue and chroma, and disregarding luminance, is processed separately by the visual cortex (and later, V4/V8) and importantly color is at a THIRD the resolution of luminance.
  • Still imagery and motion are processed by separate parts of the brain.
  • The visual cortex has a built-in sharpening filter.
  • Neurologically, resolution gets added to the scene that is being viewed over time, as the eye's saccades scan over the scene, higher resolution visual information from the narrow central foveal area is added onto the lower resolution peripheral information, improving the perceived detail of the image the longer it is looked at.

JPEG, h264, and h265 are all using 4:2:0 chroma subsampling, meaning that the color information is at a quarter of the resolution of the luminance information, which is in accordance to the human vision system which relies on luminance for detail.

They are also all using some flavor of DCT compression, which uses mathematical transforms to throw away a certain amount of spatial data.

Moving Still
Even though a movie it's just a series of individual still frames, the way data is compressed for a single still versus a stream makes a very big difference.

Because JPEG is a single still frame, all of the compression happens within that one frame (intraframe or I frame) and it's optimized for viewing as a still image. The data that is thrown out through DCT and chroma subsampling is done in a way as to be minimally noticeable as a still image.

h264 and h265 are designed and optimized for streaming motion video, so a certain number of frames are I frames (i.e.. key frames), but the rest of the frames are P frames which only store the change between the current frame and the previous frame. Because it is designed for streaming data is thrown out in a way that is minimally noticeable for motion imagery.

h265 is a newer "more efficient" codec, designed to give better compression rates for an "equivalent" perceived quality for motion video compared to h264. Nevertheless keep in mind that "better compression rates" by definition means that more data is being thrown away.


TL;DR
Motion optimized codecs like h264/h265 throw away data a little differently than a still-optmized codec (such as JPEG), by taking advantage of how human vision adjusts/adds resolution over time.

As a result, don't expect a single frame grab from h265 to look as good as a JPEG still when the JPEG still has a much larger amount of data, and uses compression techniques intended for still images.
 
Last edited:
I am not going to say that specifically, as I haven't done enough tests.

However, there is a maximum data rate, and as the frame rate goes up the data rate limits how much data is allocated to each frame.

Considering data rate only
Just shooting a static test frame with no motion at 4K, 24 FPS gives a data rate of around 100mb/s, increasing the frame rate to 48 FPS data rate goes up, but only to about 115mb/s. Twice as many frames but only 15% more data. Now this is for shooting a test chart with no motion in the frame, motion frame will change the date rate significantly. The data rate on the Mini 3 Pro is limited to 150mb/s.

If you are shooting at 60 FPS and your data rate is 150mb/s, that is 2.5mb per frame, and optimized for motion so you have key frames (I frames) and then frames that are made up of only the amount of change from the previous frame (P frames).

Taking a single JPEG is 3.8mb for that one frame, and optimized for a still image. A single raw image is 24mb. IMPLICATION: the jpeg has 16% of the original data, and the h265 frame has only 10% of the original data.

Considering codecs
Modern image data compression is based on the perceptual qualities of the human vision system (HVS), and the HVS has these qualities:
  • Contrast, edges, and fine detail are perceived from the luminance (lightness/darkness) of a scene, in V1, the first stage of the visual cortex.
  • Color, meaning hue and chroma, and disregarding luminance, is processed separately by the visual cortex (and later, V4/V8) and importantly color is at a THIRD the resolution of luminance.
  • Still imagery and motion are processed by separate parts of the brain.
  • The visual cortex has a built-in sharpening filter.
  • Neurologically, resolution gets added to the scene that is being viewed over time, as the eye's saccades scan over the scene, higher resolution visual information from the narrow central foveal area is added onto the lower resolution peripheral information, improving the perceived detail of the image the longer it is looked at.

JPEG, h264, and h265 are all using 4:2:0 chroma subsampling, meaning that the color information is at a quarter of the resolution of the luminance information, which is in accordance to the human vision system which relies on luminance for detail.

They are also all using some flavor of DCT compression, which uses mathematical transforms to throw away a certain amount of spatial data.

Moving Still
Even though a movie it's just a series of individual still frames, the way data is compressed for a single still versus a stream makes a very big difference.

Because JPEG is a single still frame, all of the compression happens within that one frame (intraframe or I frame) and it's optimized for viewing as a still image. The data that is thrown out through DCT and chroma subsampling is done in a way as to be minimally noticeable as a still image.

h264 and h265 are designed and optimized for streaming motion video, so a certain number of frames are I frames (i.e.. key frames), but the rest of the frames are P frames which only store the change between the current frame and the previous frame. Because it is designed for streaming data is thrown out in a way that is minimally noticeable for motion imagery.

h265 is a newer "more efficient" codec, designed to give better compression rates for an "equivalent" perceived quality for motion video compared to h264. Nevertheless keep in mind that "better compression rates" by definition means that more data is being thrown away.


TL;DR
Motion optimized codecs like h264/h265 throw away data a little differently than a still-optmized codec (such as JPEG), by taking advantage of how human vision adjusts/adds resolution over time.

As a result, don't expect a single frame grab from h265 to look as good as a JPEG still when the JPEG still has a much larger amount of data, and uses compression techniques intended for still images.
EXCELLENT INFORMATION and presentation!!!
 
One might also note that jpg compression removes redundant data. It doesn't just discard useful data.

If you mean redundant relative to human perception, then that's one way to look at it, but personally I wouldn't use the term redundant.

Color subsampling throws out color resolution, sampling color at about a quarter of the resolution of the luminance, because in human vision the color information is only a third the resolution of the luminance, so you can subsample color with impunity.

DCT then "simplifies" spatial detail using cosine transforms, not really an elimination of redundant data so much as a transformation to a more efficient, though less accurate, way to encode spatial data.

Also depending on how hard you squish it — you can certainly squish un-redundant data out of a JPEG, with nasty results. JPEG is lossy, so data thrown out cannot be recovered, unlike a zip file.
 
You really don't see that many good pictures from stills pulled from video.
Depends upon the video shutter speed and how much resolution you require for your intended purpose. The more jittery your 4K video, the better the still frames are for extraction. You certainly don't want cinematic motion blur in your videos used for extracting still frames. For web display purposes, and smaller prints, 4K video extracted still frames can be more than good enough. If shooting video at 30fps in sunlight without any ND filters, you can make sure you catch the decisive moment that might be missed in a shooting single frame stills in photo mode, and you can choose the best frame or multiple best frames.
 
Lycus Tech Mavic Air 3 Case

DJI Drone Deals

New Threads

Forum statistics

Threads
130,585
Messages
1,554,092
Members
159,585
Latest member
maniac2000