Bach
11th July 2002, 06:27
(or how to always get the same quality for all SVCD converted with DVD2SVCD)
Hi folks,
In the end of the last year robshot method was presented to the public as the way of extracting the maximum of quality in the conversion from DVD to DVD-R, with considerable reduction of video file size. In the following weeks, it was questioned if it would be useful to the users of SVCD too. Such quarrel was unfruitful and currently the robshots is predominantly used only by the owners of DVD-R. The results of my tests show that the method is intelligent, however have some sort of conceptual errors that makes it tiring and there are risks of image distortions. They also show that, after corrected such errors, it is possible easy use for DVD to CVD, SVCD or DVD-R conversions, and even though automatize it for the use with DVD2SVCD.
===> conceptual Errors of robshot method:
1) mpeg2 encoding must use non linear quantisation Scale (nlqS), instead of linear quantisation Scale(lqS). The values (S) of this scale are chosen by the encoder to calculate the most appropriate step quantisation of the coefficients of the discrete cosine transform (DCT), in accordance to the equation
CQ(i, j)=[alpha*C(i, j)]//[(QM(i, j)*S) ],
Where // means that the result of the division is rounded off, CQ(i, j) is the quantized term i,j of the matrix of coefficients of DCT, alpha is a numerical constant, C(i, j) is the term i,j of the matrix of coefficients of DCT and QM(i, j) is the term i,j of the quantisation matrix. When using nlqS, the variation of S is more smooth than that of lqS for S<60, so that the encoder can choose smooth values of S, minimizing the distortions.
2) the value of Q.factor must be smaller than 40. Robshot method (RoM) uses Q.factor=60, a value that is very close to the limit for which we observe distortion. Attention that the Q.factor is an exclusive concept of CCE, directly related with the quantisation, but it does not mean QM nor S, and it is not very clear the way that it is calculated. Based in the results of my tests, I believe that Q.factor is in some way proportional to the average value of the product QM(i, j)*S. Anyway, the Q.factor is the direct measure of the quality to be gotten so that, if two different films are recompressed using the same settings and the same Q.factor, can be guaranteed that they have the same quality.
3) the value of bias must be bigger than 0. If we use bias=0, the great freedom of variation of the bitrate results in the possibility of great distortions in very dark scenes. I recommend to use bias=10. It is interesting to know that this value can be directly modified in the ecl file, so that it is not necessary to wait for the end of the first pass.
4) The average bitrate, used in the second pass, is not necessarily equal to 2000. This is a crucial error. When we choose to make the first step using a fixed Q.factor, we are actually saying to CCE which quality we want to get in our encoding. In exchange, CCE will say us what is the size that the archive must have so that it gets such quality. Thus, the average bitrate used in the second pass must be calculated from the size of the mpv file created in the first pass, using the equation
avr_bitrate=mpv_size*8.192/(movie_time_length*60)
For example, if mpv_size=1,400,000 KB and movie_time_length=100 min, then avr_bitrate=1911 kbps.
5) the value of image quality priority equal to 5 is very small. It should be between 15 and 25. In accordance with CCE manual itself, When the value is close to 0,the mosquito noise at the edges (noise causing hazy part along the edges,looking like flying mosquitoes)is less outstanding, but the contouring noise (noise which looks like contour line patterns, which appear in flat and wide areas,such as dark background) is more outstanding.
6) It is totaly unnecessary to open advanced settings in order to manually modify the values of global and local bitrates. This stage of RoM can be totally discarded. When we follow instructions 2, 3 and 4, the average quantisation already is in the adequate value and the average allocation of bits already is optimized, so that any attempt of fine adjustments will be only loss of time.
In this point it is necessary to make some commentaries, for those that already usually use RoM and may insist on opening advanced settings:
= > the red regions do not correspond to images with distortion. They correspond to completely black parts of the film, that are codified with highest quantisation and very low bitrates;
= > the gray areas, if exists, already have practically the highest allowed bitrate and any attempt of change it to green will result in the creation of a new gray region in GOPs that were previously in green regions;
= > when we follow instructions 2, 3 and 4, it is verified that the value of the average Q.factor of the film is always the same, and it doesnīt matter which was the source film. This can also be verified with the use of BitrateViewer.
One question may appear: You said that it is unnecessary to use advanced settings. Does it mean it is unnecessary to make the second pass also?
No. Surely the second pass is necessary. I said that the average allocation of bits already is optimized, but the distribution of these bits inside of each GOP still can and must be optimized, and for this it is necessary to use multipass (it may be 1 or 2 pass). More than two, however, is loss of time and risk of distortion.
Now that the corrections in the RoM had been made, go to the second part of the heading of this post.
===> How to automatize robshot-Bach method:
The modifications done above make it possible to use RoM, now robshot-Bach, in an automatization tool, as for example DVD2SVCD, in order to use an adequate amount of CDs. At this moment I do it by hand, following the procedure below:
- run CCE 1passVBR, Q.factor = 30 (or 40), min = 0, max = 2500, bias = 10, and the other settings as usual;
- Calculate the average bitrate (avb_1pass) of the mpv file created by CCE, in accordance with the equation of item 4;
- Calculate the average bitrate that should be used to fill 1CD (avb1), 2CDs (avb2), 3CDs (avb3), or 4CDs (avb4):
-If ((avb_1pass - avb1)/avb1) < 0,1, then avb_1pass = avb1;
-Else if ((avb_1pass - avb2)/avb2) < 0,1, then avb_1pass = avb2;
-Else if ((avb_1pass. avb3)/avb3) < 0,1, then avb_1pass = avb3;
-Else avb_1pass = avb4.
- Execute multipass, 1, 2 or 3 passes, using min = 0, max = 2500, avr = avb_1pass, bias=10, and the remain settings as the previous one. Note that this multipass will re-use the vaf file created in the 1passVBR. So, the first pass of this multipass is NOT the vaf creation. It is the mpv creation.
Because of the way that DVD2SVCD works, the size (and consequently the average bitrate) of a SVCD depends only of the movie time length, without taking into account its compressibility. Since the author recommends that the average bitrate should be always between 1600 and 2230, the majority of the conversions done with DVD2SVCD have good quality. I say the majority, but not all. There is the possibility of, for example, producing 3 CDs of a film of 140 minutes of easy compressibility in 2 CDs, wasting then 1 CD. It could happen also that we use only 2 CDs for a film of 110 minutes when, in the truth, its compressibility is hard and would need 3 CDs.
To avoid such thing, in fact, all my last SVCDs have been done using this method.
To finish, I would like to say that this is only one small contribution for the users of the CCE, which may also be of advantage for the users of DVD2SVCD. I hope that it is of some value.
Hi folks,
In the end of the last year robshot method was presented to the public as the way of extracting the maximum of quality in the conversion from DVD to DVD-R, with considerable reduction of video file size. In the following weeks, it was questioned if it would be useful to the users of SVCD too. Such quarrel was unfruitful and currently the robshots is predominantly used only by the owners of DVD-R. The results of my tests show that the method is intelligent, however have some sort of conceptual errors that makes it tiring and there are risks of image distortions. They also show that, after corrected such errors, it is possible easy use for DVD to CVD, SVCD or DVD-R conversions, and even though automatize it for the use with DVD2SVCD.
===> conceptual Errors of robshot method:
1) mpeg2 encoding must use non linear quantisation Scale (nlqS), instead of linear quantisation Scale(lqS). The values (S) of this scale are chosen by the encoder to calculate the most appropriate step quantisation of the coefficients of the discrete cosine transform (DCT), in accordance to the equation
CQ(i, j)=[alpha*C(i, j)]//[(QM(i, j)*S) ],
Where // means that the result of the division is rounded off, CQ(i, j) is the quantized term i,j of the matrix of coefficients of DCT, alpha is a numerical constant, C(i, j) is the term i,j of the matrix of coefficients of DCT and QM(i, j) is the term i,j of the quantisation matrix. When using nlqS, the variation of S is more smooth than that of lqS for S<60, so that the encoder can choose smooth values of S, minimizing the distortions.
2) the value of Q.factor must be smaller than 40. Robshot method (RoM) uses Q.factor=60, a value that is very close to the limit for which we observe distortion. Attention that the Q.factor is an exclusive concept of CCE, directly related with the quantisation, but it does not mean QM nor S, and it is not very clear the way that it is calculated. Based in the results of my tests, I believe that Q.factor is in some way proportional to the average value of the product QM(i, j)*S. Anyway, the Q.factor is the direct measure of the quality to be gotten so that, if two different films are recompressed using the same settings and the same Q.factor, can be guaranteed that they have the same quality.
3) the value of bias must be bigger than 0. If we use bias=0, the great freedom of variation of the bitrate results in the possibility of great distortions in very dark scenes. I recommend to use bias=10. It is interesting to know that this value can be directly modified in the ecl file, so that it is not necessary to wait for the end of the first pass.
4) The average bitrate, used in the second pass, is not necessarily equal to 2000. This is a crucial error. When we choose to make the first step using a fixed Q.factor, we are actually saying to CCE which quality we want to get in our encoding. In exchange, CCE will say us what is the size that the archive must have so that it gets such quality. Thus, the average bitrate used in the second pass must be calculated from the size of the mpv file created in the first pass, using the equation
avr_bitrate=mpv_size*8.192/(movie_time_length*60)
For example, if mpv_size=1,400,000 KB and movie_time_length=100 min, then avr_bitrate=1911 kbps.
5) the value of image quality priority equal to 5 is very small. It should be between 15 and 25. In accordance with CCE manual itself, When the value is close to 0,the mosquito noise at the edges (noise causing hazy part along the edges,looking like flying mosquitoes)is less outstanding, but the contouring noise (noise which looks like contour line patterns, which appear in flat and wide areas,such as dark background) is more outstanding.
6) It is totaly unnecessary to open advanced settings in order to manually modify the values of global and local bitrates. This stage of RoM can be totally discarded. When we follow instructions 2, 3 and 4, the average quantisation already is in the adequate value and the average allocation of bits already is optimized, so that any attempt of fine adjustments will be only loss of time.
In this point it is necessary to make some commentaries, for those that already usually use RoM and may insist on opening advanced settings:
= > the red regions do not correspond to images with distortion. They correspond to completely black parts of the film, that are codified with highest quantisation and very low bitrates;
= > the gray areas, if exists, already have practically the highest allowed bitrate and any attempt of change it to green will result in the creation of a new gray region in GOPs that were previously in green regions;
= > when we follow instructions 2, 3 and 4, it is verified that the value of the average Q.factor of the film is always the same, and it doesnīt matter which was the source film. This can also be verified with the use of BitrateViewer.
One question may appear: You said that it is unnecessary to use advanced settings. Does it mean it is unnecessary to make the second pass also?
No. Surely the second pass is necessary. I said that the average allocation of bits already is optimized, but the distribution of these bits inside of each GOP still can and must be optimized, and for this it is necessary to use multipass (it may be 1 or 2 pass). More than two, however, is loss of time and risk of distortion.
Now that the corrections in the RoM had been made, go to the second part of the heading of this post.
===> How to automatize robshot-Bach method:
The modifications done above make it possible to use RoM, now robshot-Bach, in an automatization tool, as for example DVD2SVCD, in order to use an adequate amount of CDs. At this moment I do it by hand, following the procedure below:
- run CCE 1passVBR, Q.factor = 30 (or 40), min = 0, max = 2500, bias = 10, and the other settings as usual;
- Calculate the average bitrate (avb_1pass) of the mpv file created by CCE, in accordance with the equation of item 4;
- Calculate the average bitrate that should be used to fill 1CD (avb1), 2CDs (avb2), 3CDs (avb3), or 4CDs (avb4):
-If ((avb_1pass - avb1)/avb1) < 0,1, then avb_1pass = avb1;
-Else if ((avb_1pass - avb2)/avb2) < 0,1, then avb_1pass = avb2;
-Else if ((avb_1pass. avb3)/avb3) < 0,1, then avb_1pass = avb3;
-Else avb_1pass = avb4.
- Execute multipass, 1, 2 or 3 passes, using min = 0, max = 2500, avr = avb_1pass, bias=10, and the remain settings as the previous one. Note that this multipass will re-use the vaf file created in the 1passVBR. So, the first pass of this multipass is NOT the vaf creation. It is the mpv creation.
Because of the way that DVD2SVCD works, the size (and consequently the average bitrate) of a SVCD depends only of the movie time length, without taking into account its compressibility. Since the author recommends that the average bitrate should be always between 1600 and 2230, the majority of the conversions done with DVD2SVCD have good quality. I say the majority, but not all. There is the possibility of, for example, producing 3 CDs of a film of 140 minutes of easy compressibility in 2 CDs, wasting then 1 CD. It could happen also that we use only 2 CDs for a film of 110 minutes when, in the truth, its compressibility is hard and would need 3 CDs.
To avoid such thing, in fact, all my last SVCDs have been done using this method.
To finish, I would like to say that this is only one small contribution for the users of the CCE, which may also be of advantage for the users of DVD2SVCD. I hope that it is of some value.