|
Video frame deinterlacing with single field to frame interpolation |
||
| The deinterlacing of video frames is
a common problem. The old video standard has 50 fields per second.
Every two consecutive fields are combined in a frame. So we have 25 frames
per second. The fields that are part of a single frame are captured at
different time. So it is very possible that they represents a moving object
or a light source at two different locations. This causes the so called "interlaced
image". One of the best ways to overcome this problem is to use
the motion compensation algorithm. But the best motion compensation should
recognize motion, resizing, different lighting and so on of number of
complex objects. Thus the deinterlacing is a problem with complexity of the
Artificial Intelligence. By now the motion compensation algorithms often
use some kind of interpolation to restore missing pixels if the motion
vector can not be properly determined. In this situation the non-linear
interpolation (NLI) can give very good approximation of the desired result
image. In addition it is possible to
assist motion compensation algorithm with an advice of what we are
searching for in the second field. An image resolution enhancement is the
other possible area of use of NLI.
The original image, the first frame field and different interpolations can be seen on the right side of the page. The last one is the author's result using NLI. |
|
|
|
| ||
|
| ||
 Full frame restoration with bilinear interpolation | ||
 Full frame restoration with bicubic interpolation | ||
 Full frame restoration with non-linear interpolation |
