[gfmucci]

The K-7 Jpeg options for file size are 2, 6, 10, and 14 mp.

Do the 10 or 6 megapixel settings allow for any less noisy high ISO images?

If not, why not? Can someone describe the relationship between file size and noise in the K-7. I'm guessing that all photo receptors are still active in the sensor even with the smaller file sizes and the jpeg software squishes the file size down. It is not as if fewer receptors are used to reduce the heat. Am I warm? Is there technology to reduce the active receptors for smaller file sizes thus potentially reducing noise in smaller file sizes?

[falconeye]

The K-7 Jpeg options for file size are 2, 6, 10, and 14 mp.
Do the 10 or 6 megapixel settings allow for any less noisy high ISO images?

The easy answer: yes.

The correct answer: Every downsample of the full resolution (14.6MP) image to a smaller size reduces the noise at the pixel level. However, this will have no effect whatsoever for visible noise in a print. This applies to any downsampling operation, be it in camera (your question) or in post-processing. The only exception is the video mode as it doesn't use all available sensor pixels.

The exact formula is that if you reduce to 50% size (1/4 of pixels) that then pixel noise is reduced as if you would have gone to 50% of ISO. I.e., the 6 MP setting makes pixel noise at ISO1600 look like ISO1000.

[jct us101]

It's basically the same as showing your image at 50% of the full size, so the camera compresses all those pixels in the image, making the noise (that is the same) not as visible. It's a bit tricky to explain it.

[gfmucci]

The only exception is the video mode as it doesn't use all available sensor pixels.

Now that is interesting. I wonder why stills with smaller file sizes couldn't also use proportionately fewer sensor pixels (vs. compression), thus meaningfully reducing heat:noise with smaller file sizes, e.g. 6 or 10 mega pixels. For example, if only every other pixel in a 14 mp sensor were active, i.e. generating heat in the exposure process, wouldn't it be possible to reduce noise in a theoretical 7 mp file by roughly 50%?

---

Last edited by gfmucci; 04-09-2010 at 07:47 PM.

[hcc]

The easy answer: yes.

The correct answer: Every downsample of the full resolution (14.6MP) image to a smaller size reduces the noise at the pixel level. However, this will have no effect whatsoever for visible noise in a print. This applies to any downsampling operation, be it in camera (your question) or in post-processing. The only exception is the video mode as it doesn't use all available sensor pixels.
The exact formula is that if you reduce to 50% size (1/4 of pixels) that then pixel noise is reduced as if you would have gone to 50% of ISO. I.e., the 6 MP setting makes pixel noise at ISO1600 look like ISO1000.

That is correct if the algorithm to produce 12 Mp JPEG file is the same as the one to produce a 6 Mp JPEG or a 2 Mp JPEG file.

But let us remember that the JPEG is a computing method/technique to generate a compressed image file with minimum loss (JPEG - Wikipedia, the free encyclopedia). There are several standards, compression techniques and algorithms, set by the Joint Photographic Experts Group. I guess the original thread post "Do the 10 or 6 megapixel settings allow for any less noisy high ISO images?" could be re-worded as: 'Do the 10 or 6 megapixel settings use the same JPEG algorithm allowing (or not) the same relative noise, and is it the same at all ISOs?'. In my opinion this is not obvious as the compression (eg from 12Mp to 2 Mp) may allow more flexibility, yielding possibly a lesser relative noise for the smaller file size.

In addition, the answer might differ from camera to camera, with possibly a better/enhanced compression method in the newer dSLRs(?).

Has anyone some insight knowledge of the Pentax JPEG programming algorithm(s)?

Thank you in advance for any expert knowledge.

[falconeye]

I wonder why stills with smaller file sizes couldn't also use proportionately fewer sensor pixels (vs. compression), thus meaningfully reducing heat:noise

gfmucci, you're asking the right questions and just need a little bit more knowledge to become an expert yourself

In the particular case, you were right IF heat would be the #1 cause of noise and IF switching off 50% of pixels would significantly reduce heat (which both isn't the case).

Make a test: Compare noise in a shot from a cold camera immediately after power on and after an hour of continued use. The effect is measurable but not large.

For your record: #1 cause of noise is the random part in the photon particle flux from the light source via the subject into the camera lens. Nothing a sensor can influence. Google for "photon shot noise".

In the case of video: The K-7/K-x simply don't have fast enough data pipes to read all pixels 30/24 times a second. So, the majority of pixels are skipped simply. Skipped, not switches off. Video creates a lot of heat...

That is correct if the algorithm to produce 12 Mp JPEG file is the same as the one to produce a 6 Mp JPEG or a 2 Mp JPEG file.
[...]
Has anyone some insight knowledge of the Pentax JPEG programming algorithm(s)?

Be careful. You are possibly confusing two things here. The image is first downsampled (from a preprocessed raw bitmap) and second, compressed (into a file format).

I don't know Pentax' exact downsampling algorithm (it better be bicubic) but it doesn't matter for our noise discussion. The formula I gave does hold true exactly for bilinear resampling.

As for JPG compression: the algorithm is standardized. The only tuning parameter is the JPG quantization coefficient matrix (36 numbers). By selecting an image quality setting (*, **, ***, or ****) you are selecting one out of four matrices. You can inspect a JPG header in a Pentax-created JPG file to recover and inspect the matrix. AFAIK, there is nothing special about it. However, you may wish to inspect it to make sure it doesn't depend on the MP or ISO setting. It may be interesting to note though that smaller files with less pixel noise would compress better.

Summary: your comments are interesting but have no relevance with respect to the question by the OP.

[hcc]

Be careful. You are possibly confusing two things here. The image is first downsampled (from a preprocessed raw bitmap) and second, compressed (into a file format).

I don't know Pentax' exact downsampling algorithm (it better be bicubic) but it doesn't matter for our noise discussion. The formula I gave does hold true exactly for bilinear resampling.

As for JPG compression: the algorithm is standardized. The only tuning parameter is the JPG quantization coefficient matrix (36 numbers). By selecting an image quality setting (*, **, ***, or ****) you are selecting one out of four matrices. You can inspect a JPG header in a Pentax-created JPG file to recover and inspect the matrix. AFAIK, there is nothing special about it. However, you may wish to inspect it to make sure it doesn't depend on the MP or ISO setting. It may be interesting to note though that smaller files with less pixel noise would compress better.

Summary: your comments are interesting but have no relevance with respect to the question by the OP.

Thanks for the insight. It is helpful and much appreciated....

[gfmucci]

In the particular case, you were right IF heat would be the #1 cause of noise and IF switching off 50% of pixels would significantly reduce heat (which both isn't the case).

For your record: #1 cause of noise is the random part in the photon particle flux from the light source via the subject into the camera lens. Nothing a sensor can influence. Google for "photon shot noise".

Based on your explanation, this appears to be more of a S/N ratio issue not so much related to heat generation as I was somewhere led to believe. So, to attempt an analogy, in audio if you turn up the gain of a weak input, the Signal to Noise ratio will decrease, creating more distortion. With a powered antenna system, simply amplifying a weak signal received by the antenna will increase the noise. Is that what is happening with light when a processor attempts to aggressively amplify a weak signal from the sensor? Simply decreasing the number of pixels that will respond to and send an electrical signal to the processor will not reduce the distortion created by the processor in amplifying the low light signal.

I will try another analogy: The camera lens is the microphone, collecting the light instead of the sound. The photo sensors are the preamplifier, transforming the signal from the lens for the processor. In that analogy, if the pixels in the sensor were larger or more sensitive, they would give a cleaner signal to the processor under low light (low input) conditions. Am I getting close?

[falconeye]

Am I getting close?

No.

I am sorry but I think you better read background material. Thinking in analogies doesn't help.

In photography, the three sources of noise are:
1. photon shot noise.
2. read-out noise caused by the gain amplifier and analog/digital converter.
3. dark current.

Their significance, behaviour and physical cause are pretty different from each other. Try to gain an understanding of any of the three. But esp. of #1 which has no analogy in the audio world because acoustic waves aren't particles.


To give you an idea how far off you are: Think of a sensor as of a counting device counting photons. All your analogies all of a sudden don't apply.

If you really insist on thinking in analogies: a smaller image size corresponds to the typically better SNR for lower audio frequencies. Downsampling shifts spatial frequencies down.

[gfmucci]

No.

I am sorry but I think you better read background material. Thinking in analogies doesn't help.

In photography, the three sources of noise are:
1. photon shot noise.
2. read-out noise caused by the gain amplifier and analog/digital converter.
3. dark current.

Their significance, behaviour and physical cause are pretty different from each other. Try to gain an understanding of any of the three. But esp. of #1 which has no analogy in the audio world because acoustic waves aren't particles.


To give you an idea how far off you are: Think of a sensor as of a counting device counting photons. All your analogies all of a sudden don't apply.

If you really insist on thinking in analogies: a smaller image size corresponds to the typically better SNR for lower audio frequencies. Downsampling shifts spatial frequencies down.

Politics makes more sense to me. This helps explain why I wasn't a science major.

[Rondec]

It is too bad that video isn't done in such a way to maximize high iso. Seems as though it should be able to easily shoot 6400 iso since it is only taking 2 megapixel images off of the sensor.

[falconeye]

It is too bad that video isn't done in such a way to maximize high iso. Seems as though it should be able to easily shoot 6400 iso since it is only taking 2 megapixel images off of the sensor.

Nope.

The K-x sensor chip bandwidth is so small that only a small subset of sensor cell values are actually read-out (in video mode). Therefore, the pixel noise in video is at least as high as pixel noise (noise at 100% crop magnification) in still mode.

This is why the K-7 can beat the K-x in video noise. Because the K-7 has a faster chip and reads out a larger subset of sensor cell values.