Sitting staring at a line of cameras on my desk... which is the "better" camera? Shouldn't one be obviously superior for EVERY use? Why all these different cameras? HMMM... I know some of them will be "going" and a couple are "new", so I've got to take the time to see what they can really do...



I'll postulate that there are design limits for EVERY camera, such that each camera meets a need or a few needs, but one won't do "everything".

Super sharp images are "great", but then you discover the "talent" wants "soft skin" so their "reality" isn't so obviously showing. Then there's moire and aliasing...

I think the "average consumer" wants a camera that as accurately as possible "captures" the moment - the colors (maybe a bit more vibrantly than they really were, so the "memory" is more vivid), the little details, and of course most of the time great images in BAD lighing conditions... They aren't interested in "grading", funky blue green or orange "film like" color casts, or "special effects". If it fits in your pocket, even better! That's ONE "market"...

Someone fancying themselves as a "filmmaker" may have an entirely different set of "wants" - "flatter" footage that they can toy with easier in post, artsy DoF, and of course they have a lighting and sound "crew"... pocketability optional...



As I look at these cameras I'm considering for "thinning from the herd", I realize each has strengths and weaknesses, it's charms and foibles. Not a one is "perfect", at least not for "every" purpose or possible use.

It comes back to the truth that the "best" camera is the one you are comfortable using, the one you'll have with you when the "content" appears, and the one that captures a solid, technically "usable" image and sound when you ask it to. It's a tool to capture "content"... if one doesn't do it, there's probably a dfferent one that will, or if they can't do it yet, give it time (presuming you want to wait to capture your "content" rather than getting it NOW).

"Then there's moire and aliasing... "

With most of the DSLRs there seems to these artifacts. Here's a test of the NEX-5n:

"We found that the NEX-5N was able to consistently reproduce around 700 lw/ph of horizontal sharpness, and around 650 lw/ph of vertical sharpness. The camera hit much higher highs than that (it occasionally touched in the 1000 lw/ph, which is ridiculous for an APS-C camera sampling down to an HD signal), but between 750 and 950 lw/ph frequencies, the camera produced a circular banding error that was incredibly distracting."

"incredibly distracting" is what I see, that seems even worse from the 24MP NEX-7. So, I'm a bit concerned Sony will push the 24MP into the VG30 for marketing reasons.

But why no aliasing with the JVC cameras?

One thought is the 12MP 4:3 sensor is only 9MP when using a 16:9 window. In the PX10, the image is "converted" to 2MP. In the HMQ10, the image is "converted" to 8MP. Fundamentally, with QuadHD each image is simply recorded. That's exactly like taking photos at 60fps!

I suspect the DSP is simply downscaling 9MP to 2MP -- which makes me wonder if the other cameras are still dropping lines. Perhaps dropping pairs -- to keep the Bayer pattern -- but dropping nevertheless. The Canons used to drop every fifth line which is really bad.

PS: 4K2K will eliminate the need for edge enhancement (ringing) so the images will look more like film -- not less. That's why film has high resolution and yet "nice" edges.

"We found that the NEX-5N was able to consistently reproduce around 700 lw/ph of horizontal sharpness, and around 650 lw/ph of vertical sharpness. The camera hit much higher highs than that (it occasionally touched in the 1000 lw/ph, which is ridiculous for an APS-C camera sampling down to an HD signal), but between 750 and 950 lw/ph frequencies, the camera produced a circular banding error that was incredibly distracting."
I get a bit concerned when I hear statements like "it occasionally touched in the 1000 lw/ph". The resolution characteristics of the camera aren't going to vary minute by minute! It makes me wonder strongly how they were doing the measurements.......

I suspect it may have been by a method which wasn't distinguishing between true resolution and aliasing, and if all you've got is a pattern of horizontal and vertical lines that is difficult. (Hence the value of zone plates and the "trumpets" such as on the chart Adam used.

The further comment about "circular banding error" reinforces that idea. I suspect they were never seeing any "real" resolution above about 700 lpph - just aliasing - but didn't realy recognise it for what it was.
I suspect the DSP is simply downscaling 9MP to 2MP -- which makes me wonder if the other cameras are still dropping lines. Perhaps dropping pairs -- to keep the Bayer pattern -- but dropping nevertheless.
From what I hear, one of the better approaches is to not just drop pairs of lines, but pairs of columns as well. What that effectively means is that the same fundamental approach is used as for the C300 (2x2 direct read out) but of only one Bayer block in four. So the pattern goes something like:

GBGB GBGB GB
RGRG RGRG RG
GBGB GBGB GB
RGRG RGRG RG
GBGB GBGB GB
RGRG RGRG RG
GBGB GBGB GB
RGRG RGRG RG

etc, where only the photosites in bold are read. It follows you can expect little coloured aliasing, quite high luminance aliasing, and a sensitivity reduction of around 2 stops compared to what would be the case if the whole chip was being used. And resolution equal to a quarter of the total photosite count.

For 16:9 and 12 megapixel, the numbers are going to be something like 4608x2592 (=11,943,936) so you'd expect a measured resolution of about 2592/4 lpph or about 650 lpph. Which strangely enough is pretty well what we're seeing from the AF100. (It would also give an interim 1152x648 image to be scaled up to 1080 or 720 - and it's easier to scale up than down.)

Such an approach would also be expected to give a symmetrical result (equal H & V performance, unlike the 5D) - exactly as with the AF100)

The 2x2 read out approach is a very good one, but it is obviously far best used with a chip of dimensions of 3840x2160 (as the C300 does) with no need for line or column skipping. In that case there is full resolution, no need for up/downconversion, and you can expect full predicted sensitivity for the chip size.

If you use Adam's resolution testing method -- and the only one I believe in -- you slowly move the camera side-to-side to reveal those camera that use use smaller than HD chips -- which used to be very common with Pana's HD camcorders.

Therefore, one will see a changing resolution changing as the columns move across the chart. In general you can take the average -- although I suppose one could take the lowest value.

In any case -- the problem is the artifacts that they see and are present on ALL NEX video.

Bottom-line, if you do not want to see artifacts you must not choose a hybrid where high photo taking resolution is the camera's main task. That includes the VGs because they are photo cameras in a different body. That means you need a camera designed to provide the best video.

The joke on Sony and Canon and Nikon is that a camera doesn't need more than 9- to 12MP for good photos. Who prints photos anymore? They go to the web. So JVC's approach of 16:9 9MP may indeed yield softer photos than a NEX when measured, but in reality on the web where I watch them on a 1920x1080 HDTV -- that's not going to matter even after cropping. It's marketing not photo shooters that are pushing to 36MP!

To my mind the camera makers have it backward. Aim for highest video quality! Aim for video shooting ease. The E-mount is dead without the ability to have ND filters in the camera. All Panasonic has to do is drop the GH2/3 into a camcorder body and the VGs are done with. MINIMUM DOF is simply not that big a deal. (I assume there will be a better chip for a GH3.)

PS: Skipping row and column pairs is a great way to reduce data by 4X.

If you use Adam's resolution testing method -- and the only one I believe in -- you slowly move the camera side-to-side to reveal those camera that use use smaller than HD chips -- which used to be very common with Pana's HD camcorders.

Therefore, one will see a changing resolution changing as the columns move across the chart. In general you can take the average -- although I suppose one could take the lowest value.
Almost. I believe the reason for doing what you describe is to separate aliases from "real" resolution. On a static shot it's difficult to tell which is which - pan from side to side and the "real" lines move as you'd expect - aliases "ripple backwards". Much easier to see than describe!

Hence, the limit of "true" resolution is the highest value that behaves as it should. Disregard anything that "moves backwards"!
Bottom-line, if you do not want to see artifacts you must not choose a hybrid where high photo taking resolution is the camera's main task.
Fully agreed - albeit see below!
It's marketing not photo shooters that are pushing to 36MP!
I've speculated in another thread about a possible "universal" still/video camera that may break the hybrid issue you mention above. It imagines a sensor of dimensions (4x1920)x(4x1080) which is 7680x4320, or 33,177,600 photosites. (33 megapixels)

Obviously, it's capable of high quality stills. But read in the same way as the C300 it's also capable of full 4k (actually quad-HD) with relatively straightforward processing. Furthermore, by reading 4x4 blocks directly, instead of 2x2, similar processing should yield excellent 1080 directly.

"Almost. I believe the reason for doing what you describe is to separate aliases from "real" resolution. On a static shot it's difficult to tell which is which - pan from side to side and the "real" lines move as you'd expect - aliases "ripple backwards".

Exactly. The camcorders that use H green-shift and those that use H and V Green-shift provide a mix of real resolution and aliasing. Adam's tests reveal -- as you explained -- a way of seeing what the actual resolution is.

----------

I wonder if Canon has a patent on it's super-pixel design? They read-out two rows at once and so the output needs no processing to get an RGB value for a pixel.

The Sony F65 also reads-out 2 rows at a time, but with their diagonal pattern I have the sense they use DSP to calculate an RGB pixel.

The FS100 firmware has been delayed and now this rumor:

"According to multiple sources Sony is preparing a big product announcement event for early April. I don’t have the exact date yet but my sources told me there will be multiple new products for NEX and SLT (photo cameras, *camcorders* and lenses).There are so many new products I have heard of, the *NEX-F3 a new NEX camcorder* and the new 18-135mm SAM lens"

An SLT camcorder would allow phase detection AF while shooting. Perhaps with an A-mount not E-mount. Would an ND filter then be possible?

We know the wonderful F3 -- would Sony dare release an E-mount version? If so, what would happen to the FS100?

PS: Consumer Products & Services Group (“CPSG”) and Professional Device & Solutions Group (“PDSG”) to be abolished on April 1st. Could this mean the "protectionism" of the pro group will be removed letting the best designs go to market? (In Japan, a store sells all models it thinks their customers will want. There is no consumer vs pro distinction.) This might mean the far more competent pro marketing group in NYC would take over from the not so good consumer group in CA. The "One Sony" slogan may help all of us.

Japanese products are going to get more expensive. Simple $/yen ratio PLUS need to increase profits. We might get better features at a higher price.

That and the sadly sagging dollar doesn't help... but it's all relative anyway!

Though..in the defence of the Nex 5n it has considerably less aliasing and moire compared to the 7d that I have. Shot some wide material with an f-stop of 8-11 and it looked pretty good. Way better than what the Canon was doing.

For video, the inexpensive NEX-5n makes perfect sense. The photo reviewers are all very happy with the NEX-7 because it shoots fantastic photos. Sme guys are removing the OLPF to get even more detail!

There is an inherent conflict between photo and video -- with today's technology.