filter sequence

[Gib]

Could give a kind of typical sequence to apply filters ?
I found some hints reading FAQ or forum but no sum-up at all.
I understand it depends of the original video quality, but there should be usual sequences that apply 95% of time.
For example "deinterlace" should be first if needed. I also feel that its better to denoise before sharping etc..
Putting filters in the right order is key i suppose.

[Manoman455]

Yes, I always throw in filters until it looks right then I got through the filters, rearranging them so that the sequence is optimized and also gives the some what the same results as the original. I work with DVDs as source a lot so my for sure sequence is as follows.

1) inverse telecine or deinterlace ( if needed ).
2) correct aspect ratio of video.
3) clean up ( denoise, contrast, brightness, sharpen, anti-aliasing ) - make it look nice before the initial resizing.
4) run through video enhancer.
5) touch ups (adding credits, logos, watermarks, extra effects ).

I don't know it this was what you looking for but it not just clarify and I'll try to answer again.

[Gib]

Thx for ur help.

What i was looking up is hint to put filters in the right order. Especially for what u call clean up :
--> denoise, contrast, brightness, sharpen, anti-aliasing
why not AA before sharpen for example.

[Manoman455]

hmm I was just listing the thing in parentheses but that specific arrangement sounds right. The AA I use relies on an edge mask (i think) so sharpening would make more accurate AA. I could be wrong. OK I guess I can go in depth a little more, hope it helps.

1) Inverse telecine or De-interlace first because before you do anything you want the source video you are working with to be in the most "original" state possible. DVDs usually come telecined and/or interlaced. so for other filters to work their best we have to fix these. Doing so will correct the arrangement of frames and/or fields.

2) The next thing I would do is correct the aspect ratio of the video if needed. DVDs usually come as 720x480 but that isn't the actual size of the video meant to be shown. I do this with lanczos4resize and cropping (only resize width). Some say de-noise be done before fixing the aspect ratio, but I'm just stating how I do it. The reason why not to De-noise after a full resize (length and width) is the noise is larger and will not be accurately detected and removed, in my experience no visible noise is removed when enlarging then de-noising.

3) The clean up process is really what I think of as the enhancing step. Now that you have an almost perfect version of the video you can now add anything you want. But like you said why not some filters before or after others. Sometimes I do sit for like an hour rearranging filters around until they are "in opinion" the correct order. I separate the entire process into 2 categories quality and optimization.

Quality:
The first thing you want to do is remove any ugliness, or artifacts from the video. Think if this as a repair stage. Filters like De-haloing, De-noise, logo away, De-block, remove pops, etc. For the filters just mentioned the order I would do them in is De-noise, then if you have anything you don't want (logos, subtitles, pops, etc) remove them, then De-block. The reason why I don't do De-block then the removal of logos, subtitles and pops is because De-block smooths and blurs parts and that "I believe" will get in the way of other filter's accuracy. De-halo isn't always needed but I would probably put that before De-block because of the same reasons. After what i call the repair stage you are free to use filters that actually make large visible changes like contrast, brightness, color, sharpness, AA, blur, warp sharp, gama, etc. this stage is mostly what you think looks better, I can't really say which ones to use in which order because there are too many of these type of filters. Just keep experimenting and it comes naturally soon.

Optimization:
The filter orders you should watch out for. One should never make something look really good by combining a bunch of filters together to achieve it then out of nowhere add a filter to mess it up then add more filters to make it look good again. for example:

source is 640x360
this short sequence does the following:

deinterlace
Denoise
resize(lanczos3=1280x720)
sharpen
warpsharp
blur
resize(lanczos3=640x360)
sharpen

this sequence will probably not look that good and worst of all makes the rendering process take forever.

the better order for performance would probably be:

deinterlace
Denoise
sharpen
warpsharp
resize(lanczos3=1280x720)
blur
resize(precise bicubic=640x360)

notice how only one filter is used after the resizing, remember that the higher the resolution the more the filter has to do, so try to keep as many filters on the original source resolution as possible. On resizing back down choose a faster resampler because lanczos3 is not needed to scale down. I also removed the last sharpen filter as it is no longer needed. This is just an example and i don't think anyone would use either of the above sequences.

After rendering your final video it is ready to go through video enhancer.
Then if you plan on going further pop the output into a video editor like Vegas of premiere and add credits, logos, etc..

Again hopes this helps, and if some of what I mention is wrong is doesn't sound quite right please don't hesitate to mention so.

[Gib]

Thx u r cool.
Very helpy for me to have experienced users hints.

[SeeingMole]

I would recommend you to skip the step "2) correct aspect ratio of video."
This is just another word for resize.

There is no reason to "correct the aspect ratio" of the video at all.
When you resize at the end, you would automatically "correct" any aspect ratio anyway.

For example, DVD at 720x480 16:9 and you'd like to resize to 720p, there's no need to "correct the aspect" by resizing to 854x480 or 720x405 first, run filters or whatever at this resolution, then resize it to 1280x720.

Just process it at 720x480, then resize it directly to 1280x720.
Let any filters run at it's original resolution.

You should only resize the video once. Resizing several times will just degrade it.

I would recommend you to resize to your final output size as the last final step.

This is the order I would recommend:

deinterlace (remember to set the correct field order, tff or bff)
crop (crop out any black borders)
contrast/brightness/colour balance
sharpen / denoise (which comes first really depends on the source, however, I would usually sharpen before denoising)
resize

Note that it is really not necessary to use ALL of these filters for every video.
It all depends on the quality of your source. Normally I don't usually sharpen at all because it only creates an artificial sharpened image with no real additional detail. Denoise can be useful if it's very noisy.

[Manoman455]

I would keep step 2 since one might want to add filters in after the correction and re-size after rendering the video, this would work best from my experience, you can re-size to the desired resolution first then add filters but I find that adding filters before resizing (main re-size - not the ratio correction) gives better quality. Also filters work more efficiently because of the lower resolution.
I understand your statement about degrading quality from using multiple re-sizes but in my experiences resizing it makes filters work better....for example:
De-noise...when run in the regular 720x480 only removes some noise and I get these ugly blots which I then have to use something like smart smother to get rid of but then that filter also degrades quality by blurring.
lets say the original was 640x480, I re-size to that resolution and then De-noise, I get a better result, this might not hold true for you but that is what I found and have been using.
To be clear I'm not saying any of us is wrong its opinionated I guess and not every video is the same, I work with anime a lot and maybe that is why I see what I see.
Your opinion helps  .

[SeeingMole]

OK, it probably doesn't matter much for anime and cartoons.

But for regular videos, I would not recommend such resizing at all because by lower the size, you remove information which would be useful when you do the final upsize.

Of course filters seems to work better at the lower resolution, simply because there is less information to work with. When you resize down, any noise would also be interpolated down.

[Dee Mon]

Resizing down is usually done by calculating for each destination pixel weighted average of several pixels around its original location. Many spatial denoisers work by calculating weighted average of pixels around the current one but only those which differ not so much. So basically resizing down is a very simple denoising method, random noise gets cancelled out in the weighted average.