Why is video (capture) soft?

[cole t parzenn]

And why are the D-21 and Alexa relatively sharp? Even via overly-compressed internet video, film (and, to a lesser extent, the Arris) is much sharper. I'm thinking of the Reds, CineAltas, and Genesis, in particular.

[David Mullen ASC]

You have to know the details behind the image processing because you can add any degree of sharpening after capture.

 

Another factor is the strength of the optical low-pass filter in front of the sensor.

 

A third is that the F35 and Genesis are limited to 1080P RGB but use an RGB striped color array rather than a Bayer pattern, plus seem to me to use a relatively mild OLPF to gain more edge sharpness at the risk of more aliasing artifacts.

 

I saw a test posted and projected at 4K of some of these cameras (including 35mm film scanned at 6K and finished at 4K) and on the big screen, the impression of detail pretty much followed the sensor and recording specs, it went from most detailed to softest: Sony F65, Red Epic, Alexa and 35mm (tied). The Sony F65 has about 6K worth of photosites, though due to the diagonal arrangement there is 8K of green photosites horizontally, the Epic had 5K, the Alexa had 2.88K. Like I said, the 35mm version was scanned at 6K but visually seemed more like there was 3K of detail.

 

But the truth is that all of these formats and cameras when used for a movie have some degree of post sharpening, especially if finished and shown at 2K so it isn't hard to make a lower resolution camera look sharper than the higher resolution one.

[John E Clark]

And why are the D-21 and Alexa relatively sharp? Even via overly-compressed internet video, film (and, to a lesser extent, the Arris) is much sharper. I'm thinking of the Reds, CineAltas, and Genesis, in particular.

 

Film scanning has several 'benefits' that haven't been 'cheaply' available to original image capture digital devices. A film scanner may take several seconds to 'scan' the film frame, may digitize with less noise to higher bit depths, etc. Obviously at shooting rates of even the standard 24 fps, the time to 'scan' the frame is limited, which means 'fast' conversions, which in general means 'more noise', or fancier methods to avoid such noise. Other items such as subsampling 'color' rather than full 2k worth of R, G, and B, in some cases giving 2K for G, but R,B getting 1/2 or 1/4 that resolution, leads to certain results in less sharpness if there is sufficient change at a 'color edge'.

 

One 'early' method of dealing with this is to have separate R, G, B sensors, each with 'full resolution', but that is an expensive way to go, and for the NTSC/PAL cameras, usually resolution was limited to 720x486(576 for PAL), and not to mention non-square pixels, which had a 'softening' effect when viewed on 'square' pixel devices... But this method also required a beam splitter which in addition to the 3 sensors would add to the 'cost' of the device.

 

In terms of output and sharpening... sharpening does not 'add' any information to the image. What 'sharpening' does is essentially increase the contrast at 'edges' in an image. How much sharpening is dependent on the resolution of the material, and the intended display device, and usual 'viewing' distance, such that the human eye will 'integrate' to result in the perception of a 'continuous' image.

 

Much of this 'magic' has been buried in the 'production process', but with the increase of distributions which may not have the funds to pay for such services, some material shows up which does not have the sharpening 'tuned' for the ultimate display.

 

Or, someone has take material tuned for one distribution type, and simply 'subsampled' to create output for another medium.

[cole t parzenn]

You have to know the details behind the image processing because you can add any degree of sharpening after capture.

 

Another factor is the strength of the optical low-pass filter in front of the sensor.

 

A third is that the F35 and Genesis are limited to 1080P RGB but use an RGB striped color array rather than a Bayer pattern, plus seem to me to use a relatively mild OLPF to gain more edge sharpness at the risk of more aliasing artifacts.

 

I saw a test posted and projected at 4K of some of these cameras (including 35mm film scanned at 6K and finished at 4K) and on the big screen, the impression of detail pretty much followed the sensor and recording specs, it went from most detailed to softest: Sony F65, Red Epic, Alexa and 35mm (tied). The Sony F65 has about 6K worth of photosites, though due to the diagonal arrangement there is 8K of green photosites horizontally, the Epic had 5K, the Alexa had 2.88K. Like I said, the 35mm version was scanned at 6K but visually seemed more like there was 3K of detail.

 

But the truth is that all of these formats and cameras when used for a movie have some degree of post sharpening, especially if finished and shown at 2K so it isn't hard to make a lower resolution camera look sharper than the higher resolution one.

 

What's the striped color array? I'm a little surprised by the results of the test you mention. I don't think I've seen any F65 (I was thinking of the HD CineAltas) material but Reds often make actors look like life-size plastic figurines, to my eye. (Though I've never seen Red footage at 4K...) Also, while a 60 lp/mm stock in S35 is, in theory, 2.88K equivalent, the Alexa, in theory, shouldn't match it, having an OLPF and a Bayer array - right?

 

As an interesting-to-me aside, this is reminding me of seeing "August: Osage County," in 2K. It was low enough resolution that I couldn't make out the insignias on cars in wide shots but the images relayed textures in a way that I've never seen from video.