krementz Offline Upload & Sell: Off
|
p.5 #11 · p.5 #11 · Did I get it right? Canon just released another 18MP DSLR? | |
fraga wrote:
Krementz, You are right, I agree with you, I just mentioned specifically MP solely because that was what the thread was about (another 18MP camera).
High ISO has become much, much, much better.
But do you think it will get better by the same magnitude in the next 10 years (serious question, I would like your opinion).
I ask because I feel that high ISO gains are starting to become incrementally slower.
The 5DMIII and 70D are somewhat better than their predecessors in this regard, but I am under the impression that in these two specific cases, hardware is more likely the reason behind it, not exactly better sensor technology.
...Show more →
fraga, ISO has not increased as fast as other microelectronics, and I do not think it in the future, either. However, once ISO gets into the multi-100,000 range, an "exposure" may be multiple shots measured in microseconds, rather than a single exposure measure in milliseconds. Having hundreds or thousands of stacked microsecond exposures on a very large sensor should give some tremendous opportunities for noise reduction, image stabilization, and motion blur control. In dim light, the total exposure could still be say 0.5 seconds, but saved as 1,000 images. Sampling the entire stack means no gain is needed, separating them out enables motion control, and highly redundant date should enable tremendous noise control.
The traditional ISO measurement number probably becomes meaningless or incalculable, but the marketing types will "translate" it to a pseudo-ISO that sounds sexy. ISO 1 million, anybody? Perhaps multiple sensor arrays, similar to the Foveon that SIgma is working on, will be used. Already stacked images are used in astrophotography; I see this just being extended by an order of magnitude or two, with a huge amount of software processing to account for motion.
Most images, even in low light, have basically two components. One is frozen, eg, the background, rocks, floor, furniture, etc that all stay constant in relation to each other. Motion blue from the camera can be eliminated. There is some software already that does this to limited extent. The other part is the moving part, the person or football or racecar, in which most of the elements stay constant in relation to each other, and move as a group. A football stays fairly constant in shape and size, moving against the background. Isolate it, and software can then reassemble as a fixed object. More complex items, such as a human dancing or throwing a ball means dividing up the images into an arm, a leg, eyebrows, etc.
Some artificial intelligence should be able to identify specific items and keep the adjustments "reasonable". For example a face has two eyes, so never calculate a third eye in a few frames, or move the eye sockets against the skull. A hand has 5 fingers, etc. A basketball is round with a very limited amount of deflection. Load a lookup table of items with known properties to improve the algorithm. Branches may bend, bones don't, balls may deform, rocks don't, waves and flames have definable shapes, buildings are vertical and horizontal, beachballs deform more than golfballs, waterfalls flow downhill, planes fly forward, etc.
Certainly there can be images engineered that could fool the algorithm and make comic interpretations, but it would probably work for 99.99% of real images. "Funhouses" with nothing at 90 deg, mannequins made of distortable rubber, highly precise placement of objects and shapes that fool a human eye would probably also fool the electronic one, too.
So, the real ISO may not get better than 100,000, but the effective ISO may be in the millions.
It will be a fun future.
|