PhotoTool 2012

[Ags]

You've never heard of PhotoTool 2012? That's because I wrote the app this morning... :-) It analyzes Tiff files and reports on noise in dark frames and vignetting in flat frames. For example it tells me that my 55-250 zoom has 97% illumination halfway to the corners, and 76% ilumination in the corners at 135/5.0.

I'm going to use it to plan my shots, getting the best combination of speed, even illumination, and low noise. I still need to add some features, like reading EXIF data from the image, and also being able to read and graph a sequence of images (to see how noise evolves in an imaging run).

I think it would be fun to see how different cameras perform under PhotoTool analysis. Let me know if you are up for shooting some test shots or if you want a copy of the tool yourself (yes, it's free). It has no fancy user interface, it just spits out a spreadsheet. It requires Java.

[Photosbykev]

it sounds very useful for DSLR photography

[Ags]

I've been playing a bit with my new toy, and I have been looking at the field illumination of my EF-S 55-250 at 70mm.

I was looking at 2 things: (1) how much is the corner illuminated relative to the center, and (2) what is the radius of the at-least 95% illuminated field?

At F4.0:

Illumination in corner: 52% (so the corner is a full stop slower)

Distance of 95% illumination to corner: 26% (so a circle 26% of the diagonal of the chip is fully illuminated)

At F4.5

Illumination in corner: 67%

Distance of 95% illumination to corner: 50%

Af F5.0

Illumination in corner: 73%

Distance of 95% illumination to corner: 57%

At F11 (just to see if I ever get even field illumination)

Illumination in corner: 91%

Distance of 95% illumination to corner: 71%

So there is no way around it - I may as well shoot at F4.0 and take some flats! Actually, in my experience extreme flats don't work well, so I could use F4.5 which gets most of the benefit of stopping down - the circle of full illumination doubles in size for a half-point stop-down.

[Photosbykev]

even with a fast wide angle lens like the 24mm f1.4 lens stopped down to f2.8 there is significant vignetting. I don't think flats can be, or should be, avoided as they play an important part of the imaging processing. Even my 600mm f4 prime lens, at the other end of the focal range, benefits from using flats

[Ags]

Yes I need to use flats, if only for the dust that cakes my sensor :-(

But I find flats quite tricky to get right, so I like to minimize the amount of work the flat has to do.

[Ags]

I have been doing some more playing, and I have been looking at the effect ISO has on noise and quantization. There was a thread recently where there were some conflicting opinions about whether higher ISO is a help or a hindrance.

Some of the statements were:

- Higher ISO does not increase noise - it only makes the noise brighter.

- Higher ISO can help quantize weak signal better, so more nuanced digital values are assigned to low levels of light. So the image would capture more info from the faint parts, but would overexpose stars.

I used the following test protocol: I shot a dark for 5 minutes to get the sensor up to a stable temperature, and then in quick succession shot 1 minute darks at ISO 100, 400, 1600 and 6400 (increasing by 4 each time).

I adjusted my tool to report

(1) noise equivalent to the the lowest light reading possible (an integer value of 32 for my camera) at ISO100. At ISO400 the values should be stretched to an integer value 4 times higher, at ISO 6400 the values would be stretched to values up to 64 times higher... I recorded a count of this faint noise for each ISO setting.

(2) all noise was also counted - all non-zero pixels

(3) bad noise was also counted - any pixels more than 50% illuminated.

These are my results:

ISO100

\Pictures\2012-10-28\tiff\dark60s100.tiff

Lowest noise 18,311,950

All noise 7,341,908

Bad noise 1

ISO400

\Pictures\2012-10-28\tiff\dark60s400.tiff

Lowest noise 19,125,768

All noise 7,475,254

Bad noise 34

ISO1600

\Pictures\2012-10-28\tiff\dark60s1600.tiff

Lowest noise 18,049,169

All noise 6,930,629

Bad noise 7

ISO6400

\Pictures\2012-10-28\tiff\dark60s6400.tiff

Lowest noise 30,198,506

All noise 11,600,920

Bad noise 6

These are my conclusions:

I had read that ISO values higher than 1600 only multiply the digital values. If this was the case the low noise reading for ISO6400 wouldd actually drop, as half the values would be multiplied out of the range I was counting. Instead the low noise value is much higher at ISO 6400, leading me to conclude that the even the high ISO is applied in analog on the chip.

Total noise stays constant up to about ISO1600, consistent with what I have read. For some reason I get more hot pixels at ISO400 - I have repeated the test and I keep getting this result, but this is probably behavior specific to my camera. From a noise point of view, there is no reason not to use ISO1600, if you have to use short subs and you have a 1100D. ISO6400 in contrast has much worse noise, so the constant-noise rule breaks down at higher ISOs.

I had read that higher ISOs can quantize weak signals better, and that appears to be the case. Although the low noise population remains constant at ISO 100, 400 and 1600, the values are spread over more values, with values starting at an integer value of 32 for all ISOs. So the weak signal (noise in this case) was quantized much more finely at higher ISOs.

(EDIT: before someone points out that my lowest noise values are higher than the all-noise values, I count them slightly differently - for lowest noise I count the RGB values separately, while for all noise I look at the results per color pixel... I will however change the tool to report all the data the same way).

[Ags]

Hmmmm... On re-reading the above it seems a bit of a mouthful :-)

Basically, the number of pixels with noise data stays constant up to ISO 1600. The noise pixels get brighter, but the image would get brighter too.

The number of faintest noise pixels stays constant too, where 'faintest' is defined as the lowest brightness level recordable at ISO 100. At ISO 400 this is equivalent to all the pixels up to a numerical value four times higher, and so on.

Beyond ISO 1600 noise levels shoot up, with the number of noise pixels almost doubling.

There has been some debate about whether the ISO is applied by multiplying the digital values, or on the chip itself (a non-digital analog process). Based on the spread of values I would say that ISO is applied to the signal on the chip and not to the digitized values.

[Ags]

even with a fast wide angle lens like the 24mm f1.4 lens stopped down to f2.8 there is significant vignetting. I don't think flats can be, or should be, avoided as they play an important part of the imaging processing. Even my 600mm f4 prime lens, at the other end of the focal range, benefits from using flats

I have been thinking about this and I think it is not quite true that flats are a transparent solution. Taking my lens as an example, the signal in the corners is half the brightness but the noise is not, so the flat is going to double my noise toward the corners. Also, the corners will have half the quantization of the center.

A mate at work is shooting some test shots with his Nikon D60 so that will make an interesting comparison on noise data with my 1100D.