How Can I Correct This Collor-Problem? (Sample Inside)

[poisondeathray]

the field order is different in that last sample (bff)

observation: if you pick selecteven() with tgmc , frame 11 looks better than when using selectodd()

[jagabo]

The top field of frame 11 is the one with that squiggle in MJ's nose. Here's frame 11 with VirtualDub's Deinterace/Blend (left) and AviSynth's Blur(0, 1.0) (right):



They look almost exactly the same to me.

Keep in mind that both fields can have time base errors so keeping one vs. the other won't fix the problem, just move it to different frames.

[poisondeathray]

Maybe trying out different blend deinterlacers could get rid of the double images ? or maybe you can adjust the amount of blend ?

Here are some comparison shots of the same blend deinterlacer above in vdub, vs. one in after effects

[mathmax]

Thank you both for your answers

AE blend deinterlacers looks really great. Far better than vdub. Why?

[poisondeathray]

Not sure, it doesn't look to be blending anything , just blurring everything slightly

Regular avisynth deinterlacer (e.g. yadifmod +nnedi2 , even nnedi2 alone) is even better , more sharp, less aliasing

The only difference in the jitter artifacts in your earlier posts was selecting different field. Blend deinterlacers suck.

[jagabo]

AE looks like it's doing a discard field and resize, plus a blur. That would be Bob with Keep Top, or Keep Bottom, in VirtualDub's Deinterlace filter. But again, no matter which field you keep, some of the resulting frames will have horizontal time base jitter.

[mathmax]

AE looks like it's doing a drop field and resize.

what is a drop field?

[jagabo]

Drop field and resize: discard one field, resize the other. Note that I added more to that post.

[mathmax]

Drop field and resize: discard one field, resize the other. Note that I added more to that post.

oh.. that's not good because you loose half of the image this way, right?

One question: I read the avisynth documentation to improve my understanding about jitter. I wonder what is the pitch of a frame. It's described this way:

Just as in VirtualDub, the "pitch" of a frame buffer is the offset (in bytes) from the beginning of one scan line to the beginning of the next. The source and destination buffers won't necessarily have the same pitch.
The row size is the length of each row in bytes (not pixels). It's usually equal to the pitch or slightly less, but it may be significantly less if the frame in question has been through Crop.

But I don't really understand what does the text in bold mean... and why in bytes? Also is there a way to visualize the pitch of a frame in VirtualDub?

Thank you in advance for throwing light on this

[jagabo]

Pitch is the distance from the start of one scanline to the start of the next, in bytes. The length of each scanline as stored in memory is not necessarily the same as the number of pixels in that scanline:

byte position of a pixel = y * pitch + x

not

byte position of a pixel = y * image_width + x

For example a 638x480 Y plane may be stored as a 640x480 array with the last two bytes of the array unused. Ie, it's pitch is 640 even though the width is 638. This is done to keep data aligned in memory for MMX access, etc.

[mathmax]

Thank you jagabo for this explanation. This is crystal clear now

To return to the deinterlacing issue, do you think that dropping one field and resize the other is a good solution?
I personally think it's a pity to loose half of the image...

[jagabo]

To return to the deinterlacing issue, do you think that dropping one field and resize the other is a good solution?

No. As I pointed out, it's not a solution to your time base problem. For every frame that looks better with one field dropped there's another that looks better with the other field dropped.

[mathmax]

To return to the deinterlacing issue, do you think that dropping one field and resize the other is a good solution?

No. As I pointed out, it's not a solution to your time base problem. For every frame that looks better with one field dropped there's another that looks better with the other field dropped.

ok, so I'll try Regular avisynth deinterlacer (e.g. yadifmod +nnedi2 , even nnedi2 alone) as poisondeathray advised me.

On more question: I would like to get a mask with black parts when there is a lot of motion in a picture and black part when there is no or low motion. You said you could maybe use mvtools, but I don't figure out the filter to achieve this. Could you give me some hints?

Thank you

[jagabo]

Someone else suggested MvTools. I don't know much about it but I don't think it's suitable for what you have in mind.

[mathmax]

yes... I don't think it's suitable for what I have seen in the mvtools documentation.
So what would you do to create on mask per picture of my video?

[jagabo]

You can start by subtracting one frame from the next or previous to see what's changed between frames:

Subtract(last, Trim(last,1,0))

You'll probably have to do it both ways since that leaves you with 128 +/- rather than 0-255.

v1=Subtract(last, Trim(last,1,0)).ColorYUV(off_y=-128)
v2=Subtract(Trim(last,1,0),last).ColorYUV(off_y=-128)

Overlay() also has subtraction and addition functions:

v1=Overlay(last, Trim(last,1,0), mode="subtract")
v2=Overlay(Trim(last,1,0), last, mode="subtract")
Overlay(v1,v2,mode="add")

That's the equivalent of abs(frame[x-1]-frame[x]). Of course, you'll need special handling at scene changes.