Deinterlacing

What is deinterlacing?

Deinterlacing is the process of converting interlaced video (a sequence of fields) into a non-interlaced form (a sequence of frames). This is a fundamentally impossible process that must always produce some image degradation, since it ideally requires 'temporal interpolation' which involves guessing the movement of every object in the image and applying motion correction to every object.

In a CRT, interlaced display draws only half of the lines on the screen for each frame (alternately drawing the odd and even lines for each frame). This reduces flicker by taking advantage of the persistence of vision effect, producing a refresh rate of double the frame rate without the overhead of either transmitting each frame twice or holding it in a buffer so it can be redrawn.

When displaying video on a display that can support a high enough refresh rate such that flicker isn't perceivable, interlaced video can be deinterlaced for better viewing. When a display cannot interlace but must draw the entire screen each time, the video must be deinterlaced before it can be displayed. All current displays except for CRT screens require deinterlace. Though in principle there is no reason why LCD, DLP or Plasma displays should not display two fields sequentially, the requirement that half of the pixels remain black would result in a halving of brightness, even assuming that the response time of the technology could be made fast enough.

Deinterlacing method used in DynTex

There are various methods to deinterlace video, each producing different artefacts. Artefacts will always be present in deinterlaced video, as the process must attempt to combine two fields for simultaneous presentation. Any object that is moving will appear in different positions on the two fields, and simply displaying them overlaid results in a very objectionable 'comb-effect' on the moving vertical edges.  Methods often used for deinterlacing are weaving, blending, line doubling,  and some make use of motion compensation.

There is no perfect way to interpolate images in time, unless everything is moving together, as in a panned image. It is particularly difficult when it comes to deinterlaced dynamic textures: usually they are not rigid processes, and texture information can be partly lost because of spatial subsampling.

The original DynTex sequences are interlaced, in DV format. These raw sequences are provided, so that each DynTex user can perform his/her own deinterlacing method. Next to these, DynTex also offers the sequences in deinterlaced form, based on a spatio-temporal median filter (see the Database section).  The spatio-temporal median filter is a non-linear filter, and is the extension of the spatial median filter to spatio-temporal neighbourhoods. The spatio-temporal filter offers a good compromise between deinterlacing the sequence and keeping the textural information. 

Of course, feel free to use your own deinterlacing method, and let us know if you have designed a suitable deinterlacing method for dynamic textures !