Monday, June 13, 2011

Amped 5 - Deinterlace

Although interlaced video is slowly going away, analysts still see this type of evidence in their labs. The tools available for de-interlacing video can be hard to use, and even harder to explain in simple terms.

Typically, when a video is de-interlaced, the upper and lower lines are segregated and two separate files are created. The lower and upper fields correspond to separate camera views - but you're left with a choice - leave the files as is (shallow rectangle shaped half resolution) or interpolate so that the video looks "normal" (full resolution).

Photoshop's De-Interlace command can be problematic.

With Photoshop - it's an either/or proposition. Get rid of the even fields or the odd fields - then create new fields using either duplication or interpolation. ... what did you just say? .... Eliminate? Create? Ouch.

With Amped Five, it's a little different process.

Once your video is loaded, you select Deinterlace from the Edit menu. This brings up the following dialog box.

You'll want to get to full resolution somehow? How can we return to full height? We simply can't, because we actually have got a half vertical resolution file. Thus, we have to reconstruct the resulting file through interpolation. We cannot know the value, but we can guess it in some manner. In fact, in the "Interpolation" radiobox you are telling Amped Five which algorithm has to be used for interpolation:

  • "None" leaves half resolution; 
  • "Replication" copies the neighboring line; 
  • "Linear" interpolates the neighboring lines; 
  • "Cubic" interpolates by taking into account more distant lines too. 

In general, Cubic is the best choice, even though there is not much difference with the linear interpolation.

But ... now we have got all the frames of the sequence, shuffled together (as hinted at above). Often, in CCTV, subsequent frames are related to different cameras and consequently, we have many different sequences blended together. How can we separate the frames of the sequence of interest from the other ones? All we have to do is to use a selector filter, to be chosen from the category "Select Frames". 

  • "Single selector" selects just one frame; 
  • "Range selector" selects a range between two frames; 
  • "Sparse selector" selects sparse frames in sequence. 

Since the frames we need are scattered throughout the video, we need the flexibility of the "Sparse selector". We add it to the chain, then find out the initial position, and go through every frame of the video. When you encounter a frame that belongs to the sequence you need, just "Add Current" to the list. Once done, apply: you will get a short sequence, made of all the selected frames.

Going frame by frame is a pain. I'm told by the developer that the next release will tackle this problem in a more automatic fashion.

Stay tuned for more.


No comments: