Synthetic deinterlace test sequences may be completely misleading. The infamous "Cheese Slices" is a prime example. While simple in principle, such tests often do not provide textured surfaces, so field-to-field motion on repeated sinusoidal areas is not reliable. If a deinterlacer looks very clean on these tests, it usually indicates that the deinterlacer relies on some arbitrary attribute of the algorithm which may not be generally beneficial. Demeler motion compensation can be disabled, which enables a very good "bob" fall-back mode for passing synthetic tests. However, on video signals with lots of vertical detail, this bob mode flickers much more than when the default motion compensation is used. At the highest possible vertical frequency, the "1P" region on "Slices" cannot be reliably distinguished from combing effects due to motion estimation errors from occlusion or aliasing. Combing is an objectionable effect that we decided to eliminate at the expense of a slight reduction in vertical frequency response. As a consequence, the 1P region in the Slices test will be seen by the deinterlacer as all-black in one field, and all-white in the next, causing a 1/2-frame-rate flicker. With such material, it is usually beneficial to perform some vertical filtering of the input frames to the interlacer in order to mitigate this effect. The Meler interlacer can apply a vertical filter on its output prior to interlacing. However, we do not recommend using vertical filtering for "normal" video, as large regions with vertical detail very close to 1P are rare.
A subjective, but much more meaningful test of quality than "Slices" is obtained by simply looking at motion-compensated deinterlacing results on a variety of test clips. These clips should contain lots of of vertical detail, and preferably lots of vertical and horizontal camera motion. isovideo has provided many challenging sequences, with compressed originals (where available), and compressed interlaced, and compressed deinterlaced versions for your own assessment.
An objective deinterlacer quality measure is to compare sizes of lossy-compressed files (e.g. using H.264/AVC or H.265/HEVC) using a progressive video clip. In particular, it is useful to compare the file sizes of a clip that has been:
The same compression options need to be engaged throughout for meaningful comparisons. The more similar in size that (ii) is to (i), the more likely it is that the deinterlacer is doing a good job. This is because progressive compression uses motion-based matches to neighboring frames in order to compress each frame. An inferior deinterlacer may introduce visible interlacing flicker, which indicates that neighboring frames are different. Therefore, to preserve the integrity of each frame, the compressor cannot compress as effectively.
Demeler typically produces H.264/HEVC-
compressed file sizes 21% smaller than the interlaced-compressed versions.
In both subjective and objective testing,
Demeler output video
quality is usually much better than other software-based deinterlacers we have tested.
Note: In our compression tests, the Meler interlacer was used to generate interlaced video from original progressive test clips.