http://www.videohelp.com/forum/viewt...=1220821#12208 21


Nice guide Lordsmurf. Somehow I missed that one before.

The only expansion I can offer is in the area of adaptive de-interlace. Note that is refers to realtime deinterlace not MPeg encoding errors. Crude deinterlace offers two choices.

1. Toss one of the fields and repeat the remaining field twice.
This gets you a poor man's progressive image at 29.97 fps but suffers from loss of 50% of vertical resolution resulting in the stair steps on diagonals that you demonstrated.

2. Merge the fields using the "bob" technique
Best to temporal interpolate the missing lines from fields before and after in time. This seems like a great idea until something moves in the image. All motion areas go blurry.

Video engineers long ago figured a solution based on human vision characteristics. The human eye detects less detail in moving objects but is sensitive to temporal blur. So, the solution is to measure motion areas in the image by comparison with previous and subsequent frames (using both fields for fancy algorithms) and use this motion information to switch between #1 and #2 above on a pixel by pixel basis. This results in de-interlaced frame detail for low motion areas and half vertical resolution but temporally sharp (no motion blur) pixels for motion areas.

This technique is used more often than you think in live TV. Any time an image is scaled (zoomed), rotated, filtered or somehow placed into 3D space, it often needs to be deinterlaced prior to digital filtering to maintain a quality image. Of course, after the effect is generated, the output is re-interlaced into 2D TV space for the picture you see.

The same concept would apply to generating a "film look" that will be played back on a normal TV.