First generation vs. second/third generation DV video - quality decrease?

[tarrickb]

Hello,
Does anyone know if I render a DV file, outputting another DV file with exact specifications as the first (with the exception of a logo being flattened on the rendered video) if will there be a noticeable quality difference between the two?
I'm rendering these DV files using Sony Vegas, and my rendering setting is using the NTSC DV profile (with "best" selected for rendering mode).

Thanks.

[redwudz]

Usually not that much quality loss, at least out to the 3d generation, in my experience. Why not make up a short test encode of maybe a 5 - 10 minute representative video clip and run it through a few times and see how much loss there is?

I tried one with the Panasonic DV codec out to 4 generations and mostly there was a color shift towards green. There was also a little noise visible in the fourth generation. Second and third looked fine, maybe a little softer in the third generation and the color shift towards green was not as much. I could adjust the color easily enough and probably filter out the noise, though the picture would get softer. This was not from a DV source, it was one I converted with the Panasonic DV codec to DV. A higher quality source should show less detail loss and noise.

[guns1inger]

I have seen tests using the Sony DV codec across 8 - 10 generations with little degradation. Others are not so forgiving. The Microsoft DV codec and the Panasonic DV Codec are both fairly soft. For some footage the shift won't be as noticeable, but for created footage it certainly can be.

[JohnnyMalaria]

(with the exception of a logo being flattened on the rendered video)

This is an example of where our Enosoft DV Processor does something that, currently, I believe is unique.

In almost every NLE, adding a logo entails the following:

1. Decompress the entire DV video frame to either YUV or RGB format
a) Reverse the Huffman encoding
b) Reconstruct the discrete cosine 8x8 blocks
c) Transform the discrete cosine blocks to spatial domain - very, very CPU intensive
d) Re-order the blocks to create the correct frame layout
e) (Often) convert the YUV data to RGB
2. Add the logo
3. Recompress the entire video frame, creating a complete new DV video frame. Typically, wiping out any existing embedded data such as timecode, recording date etc etc.

Adding a logo with our software:

1. Identify only those parts of the DV frame that need to be changed (crudely, the region where the logo will go)
2. Decompress only those parts. But only decompress to the DCT level. i.e., don't fully decode. This saves a lot of time. Keep everything in YUV space. (Basically, steps a) and b) above)
3. Add the logo (which has already been partially encoded)
4. Recompress the new information
5. Pack the new compressed data into the original frame structure, thereby preserving all the embedded data (timecode etc etc).

Hence, the majority of the original DV frame will remain completely untouched (therefore, first generation) - only the parts that need to change will do so. All of the original embedded information remains unaffected. And, because the amount of CPU processing required is significantly less than any other existing solution that we are aware of, adding the logo can be done in realtime. If you connect two DV devices, you can add the logo to an incoming live DV feed and send it out again with the logo - without even having to capture! Although our software is still in beta, it is already in production use by a satellite provider in the US for a couple of music channels.

Once the software is released, we hope to work on a separate framework that will allow more complex DV processing (wipes, transitions etc) in realtime. This would readily scale to the HD variants of DV, too.

[tarrickb]

Hi John,
Thanks for the reply.

I think another easier solution would be to use nested projects. Using this method, all you would need to do is save a project which consists of the video and logo position. Then import this project as a nested project and the NLE (sony vegas in this case) treats it like rendered video (except no rendering has occurred).

So would it be accurate to state that nested projects eliminate the need and possible quality loss of rendering intermediate files?

[JohnnyMalaria]

So would it be accurate to state that nested projects eliminate the need and possible quality loss of rendering intermediate files?

I expect so - if that's what you want to do. i.e., use the logo'd version of a video clip as part of another project. At some point, though, the lengthy process described above will have to occur.