Auto exposure puzzle

terrypin wrote on 9/2/2014, 5:31 AM

In familiar fashion, I selected a video clip's last frame, exported a BMP and reimported that alongside the video.

It looked a bit dull so I applied Effects > Brighness/Contrast > Auto exposure to both the video clip and the BMP. But although both started with identical appearance, the BMP became much brighter than the video.

Obviously, I then went back and did what I should have done in the first place, namely applied the Auto exposure to the video clip before exporting the BMP.

But I'm curious to learn the technical explanation for the difference please?

--
Terry, East Grinstead, UK

Comments

emmrecs wrote on 9/2/2014, 8:54 AM

Hi Terry.

Just a guess on my part but the AutoExposure process on the complete clip may have been exactly that; i.e. in effect "averaged" over the whole thing. Whereas the single image has no "differences over time" as it were and hence the effect has become more pronounced.

As I say, purely a guess!

Jeff

browj2 wrote on 9/2/2014, 9:25 AM

Hi Terry,

I can't explain this. It probably has something to do with the resolution. John EB likely has the answer.

I noticed recently a feature that I hadn't used before and I don't know if you saw it. I was exporting a frame and then reimporting as are you. This feature is to right click on the video with the playback marker where you want the still, and select "Create still frame." This effectively inserts a still image at the playback marker location, replacing the remaining length of the clip with a still image. The still image created will be 1 frame long so you have to drag it out, moving everything else to the right.

This should take care of any differences in colour correcting and exposure, as the resolution would be the same.

If you want to insert it in the middle of a clip, cut the clip, go to single frame, go to the beginning of the last image and create the still frame. I think it cuts out the last image of the video clip to create the still. I tried with the playback marker at the end of the last image and the video disappeared but placed a new black video on track 2. Strange. So you have to be careful.

There seems to be a problem if I do this part way through the clip. The still image does not seem to be the image just after the playback marker. I'm still testing to find out how far along the video the image is taken from, but there is a very noticeable sudden movement from the now last video frame to the still image. If you try this, please let me know how you make out.

Regards,

johnebaker wrote on 9/2/2014, 1:54 PM

@ Terry

. . . . It looked a bit dull . . . .

I have tested exporting a single frame from different video clips - high, medium and low contrast - and get no difference in the exported jpg's and the frame's from which they were captured - all at the same 1920 * 1080 resolution ! !

. . . . It probably has something to do with the resolution. John EB likely has the answer. . . . .

Maybe .

What I have noticed in image editing software, is that when I reduce an image, say a 3840 * 2160 px image down to 1920*1080, there is some loss of sharpness / brightness / contrast depending on the resample algorithm used - normally I would use a Bicubic setting.

With MEP automatically resizing still images to fit the project resolution, and video if the source resolution is higher then the project, eg HD --> DVD, we have no control over the algorithm / method used, however I would suspect resolution reduction may play a part of the issue when there is a significant reduction in size eg HD --> DVD.

. . . . The still image does not seem to be the image just after the playback marker. I'm still testing to find out how far along the video the image is taken from, . . . .

I think that depends on where the nearest I frame is so the exact location on the clip may not be exactly where the cursor is however it should be within - IIRC ~1/8 to 1/2 second depending on the GOP structure of the video.

HTH

John

terrypin wrote on 9/2/2014, 4:13 PM

Thanks all. Very odd. I'll study more carefully in the morning (just returned from a wine tasting!) but as the resolution of the BMP was the same as the video (1920x1080), Jeff's idea looks best guess so far. I suppose it might be tested with a 'still' video.

Terry, East Grinstead, UK

Scenestealer wrote on 9/2/2014, 7:20 PM

Hi Terry

The answer to this is probably that the timeline marker was at a different place in the video when you hit Auto exposure than where you exported the still frame.

Auto Exposure adjusts for the best result where the playhead is, which may not be ideal at another point in the clip where the content or exposure has changed. It does not average unfortunately although this can be an advantage - if the change is too dramatic on an underexposed section shifting to another point that's better exposed can often give a lesser effect to the previous section.

John E.B.

That sounds reasonable but what about some of the Long GOP formats used with Phones etc., which are sometimes have up to 130 frames between I frames?

Peter

terrypin wrote on 9/3/2014, 2:09 AM

Thanks Peter, that's it!

Quite a revelation. I'd never thought about it before and suppose I'd always tacitly assumed that, for a video object, Auto exposure did base its operation on some sort of 'average'. But now that you have made me think about it logically, that's plainly impractical if not impossible. What useful result could it deliver, for example, for a room shot during which the lights were switched on or off? Or indeed my specific case, which was of a Google Earth tilt, ranging from an overhead view of dark terrain to the light sky filling the top third?

I'll ensure in future that I place the timeline marker at the start of the last frame, where the still was captured, before clicking that button.

--
Terry, East Grinstead, UK