From: Billy Biggs To: John Adcock Cc: deinterlace-discuss@lists.sourceforge.net Bcc: Subject: Re: [Deinterlace-discuss] PAL deinterlacing Reply-To: In-Reply-To: <002501c16dab$63bbc5d0$0100a8c0@adcock1>; from john@adcock8.freeserve.co.uk on Thu, Nov 15, 2001 at 07:58:58AM -0000 > This looks like a standard bad NTSC -> PAL conversion. > > [...] > > Since they try to create interpolated frames it may not be possible to > recover all the original information and convert back to 24fps. but > it may be possible just to drop all the double images if they always > only appear on one field. Ok so I did out some of the math and got these results (I'll post this soon on my webpage also.. I think I'll probably throw together a generic 'deinterlacing' page). So, I wanted to see what happens if we we apply the algorithm 'on every refresh of the PAL clock, display the current NTSC field being shown, and blur in the next one if it begins before the next refresh'. Here's what I get when I compare the algorithm's results to the 24fps original source and 3:2 pulldown-applied 60hz video (I speed up to the round 24fps and 60fps for convenience, the error is too small for these results): (discussion follows...) ms 50hz 24hz 60hz 0 ==== **** ---- 2 A A A 4 6 8 10 12 14 16 ---- 18 A 20 ==== 22 A 24 26 28 30 32 ---- 34 A 36 38 40 ==== 42 AB **** 44 B 46 48 50 ---- 52 B 54 56 58 60 ==== 62 B 64 66 ---- 68 B 70 72 74 76 78 80 ==== 82 BC 84 **** ---- 86 C C 88 90 92 94 96 98 100 ==== ---- 102 C C 104 106 108 110 112 114 116 ---- 118 C 120 ==== 122 CD 124 126 **** 128 D 130 132 134 ---- 136 D 138 140 ==== 142 D 144 146 148 150 ---- 152 D 154 156 158 160 ==== 162 DE 164 166 ---- 168 **** E 170 E 172 174 176 178 180 ==== 182 E 184 ---- 186 E 188 190 192 194 196 198 200 ==== ---- What I noticed was that regardless of if I follow the 24fps or 60fps source, the resulting output is identical (minus the loss of vertical resolution, if you're going from NTSC fields you need to do interpolation to get the spacial position right etc..). This is very interesting (and unfortunate). This pattern follows identically the results from the Simpsons page, thus giving me some hope that I've found the right algorithm. I was hoping that on the Simpsons material they actually went from the 24fps source, but as shown above, this need not be the case. I'll soon have the NTSC DVD version so I can do an actual frame-by-frame comparison. The table below is what would happen if the same algorithm was applied to true 59.94fps NTSC video. What I'm curious to hear about is if this is ever seen in practice. The results show that almost every frame is a blur between two of the original fields, which means that any screenshot you take will see motion as a ghost. Do you ever see this in practice in the PAL world? ms 50hz 60hz 0 ==== ---- 2 A A 4 6 8 10 12 14 16 ---- 18 B 20 ==== 22 BC 24 26 28 30 32 ---- 34 C 36 38 40 ==== 42 CD 44 46 48 50 ---- 52 D 54 56 58 60 ==== 62 DE 64 66 ---- 68 E 70 72 74 76 78 80 ==== 82 EF 84 ---- 86 F 88 90 92 94 96 98 100 ==== ---- 102 G G 104 106 108 110 112 114 116 ---- 118 H 120 ==== 122 HI 124 126 128 130 132 134 ---- 136 I 138 140 ==== 142 IJ 144 146 148 150 ---- 152 J 154 156 158 160 ==== 162 JK 164 166 ---- 168 K 170 172 174 176 178 180 ==== 182 KL 184 ---- 186 L 188 190 192 194 196 198 200 ==== ---- Sad tidings, -- Billy Biggs vektor@dumbterm.net