Decomb -gt; Fielddeinterlace: Threshold vs. Dthreshold like, difference, function, question

andy

this is a newbie question sorry.

I have read paragraphs about these two parameters of the fielddeinterlace function that are contained in the doc file of the decomb package. Yet, still I am not quite sure about what the difference is between the two.

AFAIU, threshold determins how 'combed' a frame has to be in order to be classified as 'combed'. Is it then that  dthreshold determines how 'combed' a frame has to be in order to get deinterlaced?

Thanks for reading

The process goes like this (if full = false): First the frame is tested to see if it is combed. This is done by detecting combed areas and seeing if there is enough combing to reasonably declare the frame as combed. The 'threshold' parameter determines the sensitivity of the combing detection in this step. Second, if the frame is declared as combed, the combed areas of the frame are again detected but this time using the 'dthreshold' parameter. The areas of the frame detected as combed using 'dthreshold' are then deinterlaced.

If full=true, the first step is omitted and all frames are declared as combed.

The next release of Decomb will allow you to directly visualize the effect of these parameters.

Now I got it.

Thank you so much for your kind reply.

Heh... go figure. I always thought dthreshold determined the strength of the deinterlacement. With a low value a very strong deinterlacement appeared over the whole picture while a high value gave a very weak deinterlacement over the whole picture.

Thanks for asking the question Malinka

BTW, is there such a thing as a quot;weakquot; and quot;strongquot; deinterlacement or are there only different ways(my technical understanding on the area is pretty much non-existant)?

Originally posted by N_F
Heh... go figure. I always thought dthreshold determined the strength of the deinterlacement. With a low value a very strong deinterlacement appeared over the whole picture while a high value gave a very weak deinterlacement over the whole picture.

Thanks for asking the question Malinka

BTW, is there such a thing as a quot;weakquot; and quot;strongquot; deinterlacement or are there only different ways(my technical understanding on the area is pretty much non-existant)?

Let me answer the second question first, then come back to your first statement. There is no strong and weak deinterlacing. For any pixel the only choice is whether it is detected as combed or not. The sensitivity of the test is determined by the threshold. Once a pixel is decided to be combed it is deinterlaced in a fixed way that does not have a strong or weak to it.

But there are different ways to deinterlace once a pixel is detected as combed. For example, you can choose blending versus interpolation. If you want to think of that as strong and weak, well, that is your business.

In effect, however, the the higher the threshold, the less of the picture will be deinterlaced, because less of the pixels will be detected as combed.

I hope that is now clear. The next version of Decomb will let you visualize it directly.

Originally posted by neuron2
I hope that is now clear.

Crystal.  Now it only remains to figure out why I seemed to observe several grades of quot;linesquot; in the same area depending on what dthreshold value I used (guess I was just imagining things...)

This also explains why I didn't quite understand everything here, I couldn't quite figure out what you meant by quot;adaptingquot;, but now it's obvious.

Originally posted by neuron2
The next version of Decomb will let you visualize it directly.

Can't wait But there are different ways to deinterlace once a pixel is detected as combed. For example, you can choose blending versus interpolation.

Donald -

Since we've got you explaining, could you expound a bit more on what you mean by blending or interpolation?

- Tom

Originally posted by N_F
This also explains why I didn't quite understand everything here, I couldn't quite figure out what you meant by quot;adaptingquot;, but now it's obvious.

In that context, quot;adaptivequot; just means that instead of deinterlacing every pixel, there is a decision based on some property, such as quot;motionquot;, or degree of combing detected. If you have an adaptive algorithm, areas that do not need to be deinterlaced are not, resulting in higher vertical resolution in those areas.

@trbarry

Suppose you have 3 pixels on successive lines and the same x offset:

a
b
c

If you decide that pixel b is combed, and you replace it with (0.5*a+0.5*c) that is (linear) interpolation. If you replace it with (0.25*a+0.5*b+0.25*c), that is blending. Note that when interpolating, the even lines are passed through and the odd lines are calculated as described. For blending, every line is calculated.  Blending thus produces a blend of the two fields while interpolation uses data from one field only. Also note that while interpolation can be calculated in place, blending cannot. Interpolation is obviously faster.

Originally posted by trbarry
what you mean by blending or interpolation?

I don't belive it.  I was on the verge of asking the same question but considered it too basic.

Originally posted by Malinka
I don't belive it.  I was on the verge of asking the same question but considered it too basic.

Heh... same thing here. I considered asking the question, but I decided not to (can't actually remember why, must've forgotten).