About making pictures   - resolution  by  Waeshael

 

These are older professional cameras with a reputation for making high quality pictures.


Newer cameras have more sophisticated noise reduction programs which mask out the effects of noise, and they seem to be be able to make noise free pictures with much smaller pixel pitch. But the noise is still there in the RAW picture. What you are looking at is something the camera has created by smoothing out the noise - by reducing color artifacts, and by damping out photon fluctuations. The image isn’t real, but is pleasing to most people who aren’t interested in high resolution (most people that view images in their browser.)


Pixel binning as a way to increase the area of the photosite

Fuji cameras use a method called pixel binning where they automatically reduce the resolution at high ISO, so that the signals from several photosites can be combined to reduce noise.

My Fuji F50fd does this - it switches from 12 MP to 3 MP and 4 photosites then make one pixel which means I can make decent looking ISO 1600 pictures and even ISO 3200 under some lighting conditions - the camera has to be kept cool (60 degrees) to keep the electronic noise low also. Click on these links for PDF files.


Fuji F50 fd High ISO Techniquesweb.pdf


NoiseAndColorAdjFD50ISO1600.pdf

Pixel size (Pixel pitch) and noise.

There is yet another benefit of lower resolution cameras. The photon noise levels are lower. Photon noise is a physical characteristic of light. Light arrives at the lens with a wobble, which is caused by variations in light intensity. This wobble prevents the electronics from making a good judgement about the level of illumination. At low light levels (shadows) the electronics may not be able to calculate whether it should make green or red light. This variability is worse when there is a low count on the number of photons hitting a photosite. If say, 100 photons are collected by a photosite, 10 of them will have different luminance than the rest (this is a law of physics). When we look at the image we say it has a signal to noise ratio of 100/10, or 10:1. that means 10% of the light created by the software is probably not the right color. For a very bright light the number of photons collected per photosite might be 10,000. The variability is calculated to be 100, and the s/n ratio is 100:1, which is very good. In the worst case where the scene is dark, the photons collected might be 20 or so, and in this case the variability would be 4, and the s/n only 5:1, which is bad.


Bigger photosites (lower resolution) give better s/n. For years the standard recommendation was for a 6 micron pixel pitch to give good s/n. In this chart here you can see that most of the cameras have a pixel pitch of 5 - 8 microns. Today many compact cameras have a pixel pitch of 2. Now, the physics haven’t changed. The bigger the pixel pitch the better the performance (here it shows the Dynamic range, but s/n follows similar curves.) Medical sensors (imaging detector) such as the Canon 9.5 MP  have a pixel pitch of 125 microns, and dynamic range is out of sight, way higher than any camera. Even with a 10 bit monitor that costs $30,000 the range of brightness is so great the image has to be “window and levelled” 1024 gray shades at a time.

If you choose to shoot TIFF or JPEG rather than RAW, you have an advantage in that you can select lower resolution and expect to get prettier pictures. If you select a resolution that is 1/4 of the maximum (4MP for a 16MP sensor), the signals from each cluster of 4 pixels is added together, so that they produce  bigger signals, and a better s/n ratio. On some cameras this can make a dramatic improvement in low light performance. TIFF captures are always 8 bits. RAW is 16 bit (actually 10 -14 bits of data but the file depth is 16 bit.) which can be converted to 16 bit TIFF for editing, then converted to 8 bit JPEG for the WEB browser.


DMC-LC5 vs D-Lux 6 pixel pitch (3.4 vs 2.1)

The higher the pixel count for a given sensor (in this case both cameras use a 1/1.7 inch sensor) means that each photosite collects fewer of the photons, and so the photon noise will be worse for cameras with higher resolution that use the same size sensor. But newer designs of sensor have ways of compensating for this. [The 1/1.7 inch sensor is  7.6 x 5.7 mm. Pixel Pitch in microns for the DMC-LC5 = 7600/2240 =3.4 microns. For the 10MP D-lux 6 it is 7600/3548 = 2.1 microns.]

But...

the push is on for more production yield from sensor fabrication shops and so sensors are smaller in area.  The APS-C sensor size is the biggest that can be made with current tooling. Anything bigger is made by passing the dies several times through the machines which causes a drop in production rate, more chances of damaging the silicon and therefore lower yield. The way to profitability for most camera manufacturers is through smaller, cheaper  cameras. And the smallest and cheapest is the phone camera. Cameras for us are just a sideline. Last year the sales of SLRs dropped significantly, while sales of phones increased dramatically. SONY is reducing the number of camera models. Canon and Nikon are selling $1500 cameras at Costco, in bubble packages.


Canon make their own sensors, Nikon uses SONY sensors. SONY make their own sensors, as do FUJI and Panasonic .

As the resolution increases on the APS-C sensor, the noise levels go up, and the dynamic range drops.


SONY Cameras and high ISO performance

SONY now has cameras that produce nice images at ISO 10,000 +. The A7S has large photosites and lower resolution than the other A7 cameras, but it can make pictures at ISO 400,000!

Just released (Dec 15, 2014) is the A7II FF camera with 24MP sensor, and I predict it will also have great high ISO images.


So what has happened that makes these sensors work so well at high ISO? Well, besides the unique sensor design by SONY, who will not release any design data to Adobe for Lightroom RAW processing - hence the colors are “off” in LR vs SONY IDC RAW converter - the image is, as usual, a creation of the camera software engineers, and the computer processor has the power to do serious image generation from a very noisy capture.

It will just take out the noise, and build an image that the software thinks should have been there. It’s as if you have a painter inside the camera who can create the picture from vague memories of the scene.

As you know, only 1/3 of the light spectrum from the scene is actually captured by an RGBG sensor, so 2/3 of the spectrum have to be “made up.” So, why not make up something to fill in the dark (noisy) areas of the scene? Nikon used to do this by sending noisy dark areas to black, so you couldn’t see any detail in the shadows.


I was at a Glenn Miller Orchestra concert last week and took a bunch of pictures on my NEX-5R at ISO 3200 using an Elmarit 35mm lens. When I got the pictures home, the red blazers of the band were more orangey that the original. I remember the colors vividly. I could not get the same color red on the computer, no matter what I adjusted. Disappointing. You see the camera cannot make saturated dark red colors. They are outside the sRGB spectrum. And the monitor cannot show any color outside sRGB (or something close.) To the left is the color profile for sRGB. You can see that the deepest reds are orangey There is nothing like firetruck red or the red of English postboxes, or the uniforms of the Guards at Buckingham palace.