ASTROPHOTOGRAPHY (What is all the noise about?) Chris Woodhouse ARPS FRAS

Transcription

1 ASTROPHOTOGRAPHY (What is all the noise about?) Chris Woodhouse ARPS FRAS

2 Havering Astronomical Society a bit about me living on the edge what is noise? break noise combat strategies cameras and sensors questions

3 A bit about me

4 living on the edge counting photons: portrait EOS CMOS sensor 1/2,000 second exposure, ISO 250 face has pixel value of 40,000 about 80,000 photons 1 photon every 6 nano seconds

5 living on the edge counting photons: deep sky KAF MP CCD sensor 20-min exposure nebula has pixel value of 65 about 30 electrons, or 50 photons 1 photon every 24 seconds ~4000 million times dimmer

6 SNR - portrait vs deep sky SNR = Signal to Noise Ratio portrait 1/1000s SNR = 200 (almost 100% shot noise) narrowband 1200s SNR = 3 (20% shot noise, 80% sensor noise)

7 M63 Sunflower Galaxy

8 what is noise? Noise is everything in an image we don't want. Light has Noise light pollution (unwanted background) + shot noise (random noise) deep sky (wanted signal) + shot noise (random noise) Sensors have Noise pixel variations (pattern) read noise (pattern and random) thermal noise (average and random)

9 light has noise Photons are like raindrops; random in nature. If you measure and compare the accumulated amount over equal areas (pixels) you will find: The randomness (noise) between pixels increases with the average amount, defined by: shot noise = mean pixel value Compare a bright scene with a dim one: 40,000 = 200,1/200th of mean value 400 = 20, 1/20th of mean value

10 light pollution and noise Individual RGB filters have a noise advantage over RGB sensors; they reduce shot noise by reducing yellow light pollution. (As do LP filters) the red and green exclude yellow (@ 590 nm)

11 noise combat strategies take up fishing add more exposure calibrate the image files find a darker site cool the sensor image processing add even more exposure

12 ripples in the space time continuum

13 add more exposure Random noise becomes less significant with more exposure. There are two similar strategies: 1. lengthen each exposure (but do not clip) 2. take more exposures and average them In essence, choose an exposure that just does not clip and take multiple exposures. Each time you double the exposure count, the averaged image has 40% less noise.

14 more exposure = better images M31 Andromeda Galaxy: 4 hours luminance 4 hours through separate R, G & B filters 8 hours total over several weeks

15 M31 Andromeda Galaxy

16 even more exposure M13 Globular Cluster: 130 x 5-minute exposures through separate R, G & B filters. ~11 hours over 3 weeks

17 M13 Globular Cluster

18 more is not enough IC1396 Nebula Shot with SII, Hα and OIII filters 120 x 20 minute exposures SHO 60 x 5 minute exposures RGB 45 hours over two months

19 Elephant Trunk Nebula- enhanced color

20 sensor calibration Why? If you average thousands of image files, you will still have sensor pattern noise, hot pixels, dust shadows and vignetting in each image. This 'noise' is constant but still annoying. Calibration uses dark frames, zero-exposure frames and images of featureless T-Shirts to create an image of pattern noise to subtract from each image and normalize with another, which makes all pixels behave the same.

21 sensors - require calibration calibration typically requires: 50 averaged zero-length exposures of nothing 50 averaged exposures of nothing at image exposure time and temperature 50 averaged exposures of a flatly lit uniform subject 50+ exposures of the image itself calibrated image = (image - dark) x (normalized flat)

22 Break

23 Astrophotography

24 welcome back

25 Rosetta Nebula

26 deep exposure and dynamic range increasing the exposure count increases the dynamic range example: Orion Nebula - particularly tricky 10 hours Hα in 30, 120, 300 second exposures 8 hours SII in 120, 300 second exposures 8 hours OIII in 120, 300 second exposures 4 hours RGB for stars acquired over 10 nights (due to low altitude)

27 Orion Nebula

28 dynamic range > 25 stops Sensor noise also reduces the sensors dynamic range but averaging multiple exposures improves it: e.g. CCD full-well capacity = 25,000 electrons, read noise = 5 electrons >> effective dynamic range i O s: 25,000/5 = 5,000 levels or just 12 bits but averaging 50 exposures increases that by 50x to 18 bit and then can be processed as a 32-bit image

29 image integration (an aside) A simple average of the calibrated images may show traces of aircraft, satellites, cosmic ray hits and meteors. While not noise, they are still annoying. When averaging, statistics sample equivalent image pixels in every image and reject individual pixels that are very different from the average value.

30 Whirlpool Galaxy M51

31 find a darker site A darker site has several benefits: less light pollution less shot noise contribution from light pollution less sky gradients same image quality with less exposure benefits visual astronomy too still has mosquitos no Internet?

32 cool the sensor Thermally generated electrons accumulate randomly during each exposure... average noise level is proportional to duration rate is proportional to temperature randomness increases with average level, like shot noise. the rate halves with ever 5 6 C reduction.... equivalent noise reduction requires averaging double number of exposure frames.

33 Horsehead Nebula, Natural Color

34 image processing image process is not the cure, but it helps; noise reduction tools (PixInsight) MURE MMT / MLT TGVDenoise SCNR - specific colors Cosmetic Correction

35 MURE Denoise (script) easy to use and remarkable: requires gain, read noise and # of images in the stack with no loss in resolution Before After

36 Heart Nebula SH2-190

37 CCD vs. CMOS sensors CCDs were more linear, less noisy but more expensive than CMOS sensors. security and scientific users still use CCDs 8 mpixel KAF8300 CCD is 15 years old cameras have mostly abandoned CCDs and spent their resources improving CMOS sensors over the last 15 years.

38 CCD vs CMOS sensors CMOS sensor architecture has faster readout speed and frame rate high frame rate encourages use of "lucky imaging" specifications are alluring but you need to be careful gain-setting changes alter dynamic range and noise caution! rapid CMOS development for astro is creating hardware and software reliability issues

39 M45 - Nebula 35 hours exposure

40 sensor parameters datasheet values: read noise (e - ) well depth (e - ) pixel size microns sensor size (mm x mm) ADC 12-, 14-, or 16-bit gain (e - / ADU) dark noise rate (e - / sec) quantum efficiency (%) linearity inferred Values - more meaningful: dynamic range (well depth / read noise) full well / area (FWD / pixel area) noise / area (read noise 2 / pixel area)

41 M27 - Dumbbell Planetary Nebula - 35 hours of exposure

42 sensor comparisons Sensor ADC Dynamic Range full well e - /area read noise e - /area dark QE Ha % KAF ICX ICX KAF Atik Horizon EOS 60Da ISO ** ~ ??? < 30 IMX ** ASI ** 0.024

43 typical CMOS architecture 1 e - read noise 10, Gain (2) x16 Output Note: Well depth > ADC resolution Read noise < ADC resolution (at low gain) Read noise > ADC resolution (at gains higher than 3) pixel well 12-bit ADC 16-bit output

44 M33 - Triangulum Galaxy

45 traits of CMOS sensors **12-bit ADC limit dynamic range and increase read noise at low gain settings due to quantization noise. as gain (or ISO) is increased dynamic range reduces e- / ADU decreases noise / area reduces full well / area reduces less quantization noise contribution Amp Glow and associated shot noise Random Telegraph Noise - Cosmetic Correction

46 gain settings - Atik Horizon camera has very different characteristics at different gains; leading to composite images: deep sky (long exposure of dim nebula) gain = 30, 512 effective dynamic range, 0.08 e/area colorful stars (short exposure) gain = 1, 5330 effective dynamic range, 0.94 e/area >> process both and combine to give non-saturated colorful stars

47 Eastern Veil Nebula

48 QUESTIONS?