You possibly noticed that in many movies - especially shot with home camcorders - in episodes where fast movement is recorded, the image looks blurred and stripped. This is because many cameras do not record 25 frames per second, with resolution 720 to 576 pixels - but 50 frames per second, with resolution 720 to 288 pixels: first the even lines are recorded, then the odd ones.

Hence, each frame you observe on your monitor is actually comprised of two half-frames, recorded with a 1/50 seconds interval and connected with each other line by line - that is, the first line of the first half-frame - the first line of the second half-frame; the second line of the first half-frame - the second line of the second half-frame; etc.

As a result, if the frame contains fast enough movement, then on the second half-frame the moving object has already moved to slightly different position compared to the first half-frame, and at the connection points the "stripes" effect occurs.
Deinterlace is a process which allows to avoid the undesirable effects while connecting the two half-frames.

Many algorithms attempt to interpolate the odd lines by the even lines' points. Unfortunately, techniques of that kind result in losing up to 50% of the initial video information - that is, half of the efforts you made while shooting your video were in vain.

Our algorithm works in a principally different way. It is based on the motion estimation algorithm taken from MPEG techniques. In our product the latter is used in attempts TO FIND the moved objects and restore the frames without losing any original information.

To speed up the process, the evaluation of movement is conducted in two steps: first the raw, then the fine positioning. Besides, the filter utilizes the SSE extension. Currently, our filter works in a real-time mode even on relatively weak computers - and on Athlon XP 1600+ it produces 35 frames per second.
In our realization motion estimation has 2 step for search.