Interlaced video

Dealing with interlaced video requires to understand how video is interlaced, then it isn't too hard to de-interlace video and get accurate digital picture.

Why video is interlaced

A movie is a sequence of images (called frames) displayed at a rate of 24 per second. PAL TV is video displayed at 25fps (frames per second) and NTSC TV at 30fps.

Video at 25 or 30fps is enough with human eye properties but on TV screen, image is perceived flickering. Engineers found that displaying the same frame in two parts (called fields), meaning one field is for the odd lines (1, 3, 5, 7...) and one field for the even lines (2, 4, 6, 8...) and doubling the rate (60 1/2 fps and 50 1/2 fps) avoid flicker thanks to screen remanence.

Therefore TV is interlaced video: 60 half-images per second for NTSC and 50 half-images per second for PAL.

More details provided by the ATSC (Advance Televison Systems Committee) at Basics of Video.

How video is interlaced

Film interlacing is called Telecine. A very good explanation is proposed by Nicky's guide at NTSC, PAL & Interlaced Explained. In Telecine, to make 30fps NTSC from 24fps FILM, the process used is called 3:2 pull-down and to make smooth video, the fields from a picture sequence A B C D are interlaced this way AA AB BC CC DD. Two frames among five are made with two fields from two different pictures.

So if you capture AB frame, odd lines come from picture A and even lines from picture B, the two pictures have a time shift of 1/24 second. Therefore object moving have their odd lines and even lines shifted and it looks like:

And if scenes change on an AB or BC frame, see how it looks like with Example 1 :

I haven't found literature about PAL Telecine, it could result with AA AB BC CD DE.

Interlacing could come from CCD camcorder or PCTV frame grabber too, I haven't found any white paper from manufacturers but if you capture a progressive source and get interlaced video, your grabber is interlacing. A progressive scan (from brand new or professional camera) is a non-interlaced frame.

Actually, interlacing differs regarding video source : films, camcorder CCD or PCTV frame grabber.

Accuracy of digital video

Here's a table of standard video parameters:
 
  NTSC PAL SECAM
images / second 29.97 25 25
ms / image 33.37 40 40
lines / image 525 625 625
aspect ratio 4:3 4:3 4:3
interlace 2:1 2:1 2:1
µs / line 63.56 64.00 64.00
digital resolution 640x480 768x576 768x576
more lines make video interlaced 240 288 288
CIF resolution 320x240 352x288 352x288
CCIR 601 luminance resolution 720x485 720x576 720x576
CCIR 601 chrominance resolution 360x485 360x576 360x576
Color subsampling 4:2:2 4:2:2 4:2:2

Note: vertical resolution is less than 525 or 625 lines per image because some are used for blanking (synchronization). CIF is Common Intermediate Format, noninterlaced, every pixel carries Luma and Chroma, this format is less demanding because 1/4 of full size but it carries very accurate data. Yes, NTSC isn't 30fps but 29.97 !

So we have Luma coded with 720x576 pixels (NTSC 720x485) and Chroma with 360x576 (NTSC 360x485) therefore every line carries meaningful information and dealing with interlaced video makes sense (versus CIF).

How to de-interlace video

Video editing software offers filters to convert interlaced into progessive video: telecine 3:2 pull-down, telecine 2:1, field swap, blending, interpolation.

Filters for VirtualDUB :
 
Filter Web Interlacing Comment
deinterlace - area based v1.2 - Gunnar Thalin Go To -> any  
deinterlace - PAL movie v1.1 - Gunnar Thalin Go To ->    
deinterlace (internal)      
deinterlace - Auto PAL movie 2.0 - Simon Walters Go To ->    
field swap (internal)     for dummy frame grabber (like PCTV)
smart deinterlace (2.4) - Donald Graft Go To ->   deinterlace only in motion area
telecide (1.1 beta 1) - Donald Graft Go To ->    

Blending fields is often highlighted as the best choice but I disagree because it blends time shifted information which can be totally inaccurate (with Telecine 3:2 when a scene change to another - like my Example 1 above with BC frame) and which results in blurring the picture. Interpolation is better to keep image sharpness. By the way, when interlacing is very dummy blending is the best solution (in the table of examples below, blend appears a lot).

I suggest to first take time to understand nature of interlaced video then to select the most appropriate filter, particularly with Telecine, and to do some testing.

Examples

I have a couple of video sources connected to my PCTV: Hi8 camcorder, standalone DVD, VHS camcorder, terrestrial TV with French channels (TF1, France 2 , France 3, M6) and satellite TV broadcasting news, cinemovies and sitcoms.

I will try to fill this table from time to time (all captures are 768x576 uncompressed):
 
Source Standard Program Channel Nature Best Filter Other Filters
Hi8 camcorder 400x400     PAL interlacing smart (phase shift) + area based  
Standalone DVD 720x576 US Movie   Telecine telecide (field swap)  
VCR (VHS) 250x240          
Terrestrial TV 720x576 News France 2 Progressive field swap + area based field swaps only
  720x576 News TF1 Progressive field swap + area based  
  720x576 US News TF1 Progressive field swap + area based  
Satellite TV 720x576 French News Canal + Unknown field swap + smart (blend) field swap + area based
  720x576 French News LCI Unknown field swap + area based (blend) deinterlace (blend)
  720x576 French Sport Canal + Unknown field swap + smart (blend) field swap + area based (blend)
  720x485 US Sport Canal + Progressive field swap + smart field swap + area based
  720x485 US News CNN Telecine telecide (field swap)  
  720x485 US sitcom Comedie Telecine telecide (field swap) field swap + area based
  720x576 French Cinema Cinecinema Telecine telecide (field swap)  

I guess this table may help me and others to understand better how to deal with interlaced video.

Bibliography

Analog and digital video: Interlaced video: De-interlacing filters for VirtualDUB: Other software :



Top | Back to DV-NLE