Darkness around corners

[Mephesto]

http://neuron2.net/hotspot/hotspot.html

So I just fixed a home-made video that was shot with a modern camera with smudges around the perimeter, concentrated in the corners -- probably due to the owner wiping the lens clean with a napkin/rag and pushing all the crap to the corners.

It isn't really visible until deshaking. I fixed it with several applications of the hotspot filter for VDub but it was inconvenient to work with and didn't really fix it perfectly because I had to create a manual mask but it was good enough.

What do you gurus use?

In severe cases (like in scenes in Blair Witch project shot with a flashlight-equipped camera in the dark) I would blur all the frames with a high radius then subtract the normal frames with the blur but I hate doing this because it flattens the intensity and makes everything look unnatural. Correcting the luma would correct it for one thing (human skin) then screw it up for another (rivers, skies, grass).

[poisondeathray]

Is it only darker , like a vignette ?

But you mentioned "smudges" - so is it blurry as well was darker ?

What are the temporal characteristics ? If it's "crap" pushed to the corners was it something on the lens, but stationary? (thus only the camera movement causes it to "move") . Or did wiping or something else occur and the characteristics change over time ? Or is it completely something else ?

Why isn't it visible pre deshaking ?

[Mephesto]

Oh man, when I read my own post I knew I knew how I would respond if the roles were reversed. I would be far less nice and ask for screenshots right away before I'd help.

I admire how you try to communicate as clearly as possible while being cautious about asking for photos to not be a dick.

It's not really blurry, just darker, possibly slight contrast change. It's stationary.

Screenshots, one original one deshaked:

[poisondeathray]

I'd probably use an overlay with feathed mask - ie. composite a brightened version over the corners . It's the same idea as hotspot (I think, never used that vdub filter)

Why after deshaking are the dark corners moved in? Did you have some edge compensation setting like mirror edges something like that ? When you have zoom for compensation, it might even get rid of the vignette!

[Mephesto]

I'd probably use an overlay with feathed mask - ie. composite a brightened version over the corners . It's the same idea as hotspot (I think, never used that vdub filter)

Aaah what was I thinking. Fix that one frame in photoshop, subtract source frame with fixed and subtract the video with the subtraction.

Problem is, the entire perimeter is slightly darker not just the corners but the corners are most visible. I couldn't manually fix the non-corner sides but they aren't that important so meh.

Why after deshaking are the dark corners moved in? Did you have some edge compensation setting like mirror edges something like that ? When you have zoom for compensation, it might even get rid of the vignette!

Because this video is extremely shaky. It was zooming in too fast and the deshaker tried to chill it out a little by un-zooming that frame that rapidly zooms and filled the corners with extrapolated data from previous and future frames, but since the part of the wall on the other frame that it used to extrapolate was not in the corner, it's not smudged so the deshaker isn't smart enough to get the clue that corners are always smudged and should be darkened if it isn't.

Same reason why a shaky video with a logo when deshaked will be a video with a dancing logo. My video has dancing corner smudges.

EDIT: What I wanted to ask was, is there a tool that can average out all the frames in the video to one so stationary anomalies can be properly and precisely isolated?

[poisondeathray]

I'd probably use an overlay with feathed mask - ie. composite a brightened version over the corners . It's the same idea as hotspot (I think, never used that vdub filter)

Aaah what was I thinking. Fix that one frame in photoshop, subtract source frame with fixed and subtract the video with the subtraction.

huh ? that's a few too many subtracts

Subtract doesn't really work for this type of thing, and you would have to manually fix it per scene using that method

I would use a feathered luma mask , with a brightened version of that video. This way you should only to do it once instead of doing it by scene .

Problem is, the entire perimeter is slightly darker not just the corners but the corners are most visible. I couldn't manually fix the non-corner sides but they aren't that important so meh.

Same thing; you make the mask to cover it. If it doesn' t move, you should only have to do it once

EDIT: What I wanted to ask was, is there a tool that can average out all the frames in the video to one so stationary anomalies can be properly and precisely isolated?

This this of approach doesn't work very well, unless you aren't describing what you had in mind very clearly... or it's for a special type of static clip (like tripod shot, no motion)

[Mephesto]

huh ? that's a few too many subtracts

lol, I dunno how to overlay with avisynth so I use Subtract() to do mah bidding. This is how I add chroma from one video to another:

equalized=imagesource("C:\tealgrey1.png")
original=imagesource("C:\teal.png")

b=original.greyscale()

chroma=subtract(b,original)

subtract(chroma,equalized)
invert

Same thing; you make the mask to cover it. If it doesn' t move, you should only have to do it once

How do you do feathered luma masks on avisynth?

This this of approach doesn't work very well, unless you aren't describing what you had in mind very clearly... or it's for a special type of static clip (like tripod shot, no motion)

There is lots of shaking motion but the entire video is one scene in the same room and with the exception of zooming into the couch, the upper corners are always the color of the wall: blue-green.

I figured a photographic average of the entire video would isolate the smudges because they are in the same place so the upper corners and the perimeter would end up being slightly darker than the rest of the frame if a composite average was achieved, am I wrong?

[poisondeathray]

lol, I dunno how to overlay with avisynth so I use Subtract() to do mah bidding. This is how I add chroma from one video to another:

equalized=imagesource("C:\tealgrey1.png")
original=imagesource("C:\teal.png")

b=original.greyscale()

chroma=subtract(b,original)

subtract(chroma,equalized)
invert

so does it work in this case ?

How do you do feathered luma masks on avisynth?

You do the mask feathering in photoshop / gimp / or image editor . Areas 100% white are 100% transparent, shades in between are partially transparent .

You have original video on bottom. Brightened version on the Top. Mask in the middle. You only want to affect those regions on the edges, so you make your mask accordingly (it's much easier since it's a static defect) . Think this overlay as "covering up" those affected areas with a brightened version of the video . This mask will be black in the center, so the center will be unaffected - only the edges will be brightened

o= original video
b= bright video
m= mask
overlay(o, b , mask=m )

There are a few more tricks , like opacity and blending modes, and many manipuations you can do on the "bright" video to make it blend better

This this of approach doesn't work very well, unless you aren't describing what you had in mind very clearly... or it's for a special type of static clip (like tripod shot, no motion)

There is lots of shaking motion but the entire video is one scene in the same room and with the exception of zooming into the couch, the upper corners are always the color of the wall: blue-green.


I figured a photographic average of the entire video would isolate the smudges because they are in the same place so the upper corners and the perimeter would end up being slightly darker than the rest of the frame if a composite average was achieved, am I wrong?

That makes it easier, it's basically 1 scene and it sound like a background replacement (with the exception of the 1 zoom scene) ? . Then it depends really on how you take the average (over what time period), and over what areas. I'd have to look at a clip

The overlay approach works (mostly) for different scenes, colors, lighting (unless there is clipping), different objects, zoom ,movement etc.... (as long as the defect is static and the mask "fits") - it's a more general purpose approach

[budwzr]

That looks like vignetting from the lens in wide position, or a conversion wide angle lens. Smudges would not be perfectly uniform like that.

[jagabo]

That looks like vignetting from the lens in wide position, or a conversion wide angle lens.

Most cheap film projectors have the problem too.

[Mephesto]

so does it work in this case ?

Yes but I didn't spend a lot of time correcting the corners in photoshop so it wasn't perfect and the darkened corners are not 100% static, they are a bit worse later on in the video so there's no one perfect mask. The three masks I created in Hotspot are a good balance but it was annoying making them.

You have original video on bottom. Brightened version on the Top. Mask in the middle. You only want to affect those regions on the edges, so you make your mask accordingly (it's much easier since it's a static defect) . Think this overlay as "covering up" those affected areas with a brightened version of the video . This mask will be black in the center, so the center will be unaffected - only the edges will be brightened

o= original video
b= bright video
m= mask
overlay(o, b , mask=m )

There are a few more tricks , like opacity and blending modes, and many manipuations you can do on the "bright" video to make it blend better

I see how it is but the hard part is that the darkening in the corners is not uniform so the feathered mask will have to be accurate.

That makes it easier, it's basically 1 scene and it sound like a background replacement (with the exception of the 1 zoom scene) ? .

There are many many zoom scenes. The cameraman was a moron who didn't look like he knew what was doing. The woman in the shot even got that weird look in her eye when she heard the zoom buzzing sound on the camera for the 20th time. She's giving a lecture, answering a few questions and this guy keeps zooming in and out on her like she's a porn star. And because he can't hold the camera steadily, the zoomed in parts makes the shakiness much worse.

Then it depends really on how you take the average (over what time period), and over what areas.

The entire clip. Like http://jk-keller.com/daily-photo/2003/
Sometimes in bored moments I wondered what a composite average of an entire movie would look like but this is the first time I have actual use for such a technique. I don't think the areas would matter because the smudges are almost completely static.

The overlay approach works (mostly) for different scenes, colors, lighting (unless there is clipping), different objects, zoom ,movement etc.... (as long as the defect is static and the mask "fits") - it's a more general purpose approach

I'll try this next time I run into this problem but I'm not convinced this is more convenient than the masks I had to create for Hotspot. The only difference was the feathered masks were created by drawing a solid mask over the affected area and blurring it to oblivion to create the feather. Had to adjust it 50 times before it was perfect and ended up doing left and right side separately.

That looks like vignetting from the lens in wide position, or a conversion wide angle lens. Smudges would not be perfectly uniform like that.

The clip is 4:3 and the smudges are 75% uniform, they do look worse later in the clip.

[budwzr]

When the lens moves toward telephoto(zooms), it goes away right?

[Mephesto]

When the lens moves toward telephoto(zooms), it goes away right?

No, gets a little worse during zoom but it's harder to tell because the couch and her clothes are not a solid color like the wall.

[poisondeathray]

I'll try this next time I run into this problem but I'm not convinced this is more convenient than the masks I had to create for Hotspot. The only difference was the feathered masks were created by drawing a solid mask over the affected area and blurring it to oblivion to create the feather. Had to adjust it 50 times before it was perfect and ended up doing left and right side separately.

The difference is you're sampling from the same frame (to apply to each & every frame) with the overlay.

The reason why averaging doesn't work well - If you average over , say even 3 frames, the difference in brightness will make it not match up. The larger the time horizon for the average, the more it will likely not match up

With a static background scene you can even sample from a few pixels over (eg. crop and resize width), so the adjacent pixels from the same frame cover up the defect, and composite and blend it better with a mask

It's easier to do in something like after effects, because you can do the mask and get feedback right away, re-adjust . (You don't even need a luma matte, you can do the mask and feathering directly)

[budwzr]

So what PeedeeArr () is saying, to keyframe the mask as needed, should work, but you need an NLE to see what you're doing.

Make a radial ambient occlusion and eyeball it every few seconds.