HOW IT's DONE -IMAGE PROCESSING

HOS  filter image

Filter changer - 

1L/2S/3H/4O/5 clear

30s crude align 

120s needs fine align.

Calibration frames

Bias - readout noise - lens cap and short exposure

Dark - system noise - lens cap and same exposure time as lights (images)

Flats - vignette - white tee shirt over lap top screen.

Dark flats -noise on flat - lens cap with same exposure time as Flats

Collect at least 10 long exposures, with dark flats.  1/4 and 1/10 times for HDR.

Stack images in Deep Sky Stacker - group the different filters and exposure times, stack repeated images  using a common reference selected from best H image. Take non repeating images, use same reference frame, check images in turn and save. Use separate groups for different image sets that DO NOT include reference. 

In PS assemble HSL layers, change mode to RGB, align, equalize core color area, paste into RGB layers. reset black level, separately merge HDR color images, then merge color and L shorter exposure times.

Exposure times 

f2 30secs ISO2000 RGB is benchmark for good nebula =  

Filter increases exposure 500x, reduces bkg 6x.   

G30 is ISO 6000 so f2 240s G30 HOS should resolve M8. 

Example Rosette Nebula HOO image   f2  120secs x 4 frames per channel.  In Wimberly S/N  for Nebula 2.5:1, a 2x improvement with line filters for a M5.5 nebula, background 4 AU. Eagle nebula (pillars of creation M6.0 should increase exposure time to 200 secs. The nebula had a intensity of 10 AU, so darker Bortle or longer exposure for dimmer objects. 

Visualize corona 

Exposure conditions http://eclipse.gsfc.nasa.gov/SEhelp/eclipsePhoto.html

Example from 2016  1/60 middle corona, 1/20 upper corona. 1/3 outer corona ISO500 1000mm on NikonP1000. Need to be on tripod and need much shorter exposures, use Sony for lower noise. Ay ISO500 1/400 inner corona, diamonds 1/1000, prominences 1/2000 ISO100, chromosphere 1/5000 ISO100, Baileys beads 1/5000 ISO50.

At ISO100  6 levels - 1/2000 - 1/400 - 1/80 - 1/16 - 1/3 - 1.5 secs prominences to outer corona.  

Use Nikon for prominences at max. - video for image average. 0.6 degree field  ISO100.

Use Sony on Orion ED80 for corona because solar filter is easy to add. Time lapse  video up to totality and then exposure series, back to video for diamonds. 

PS function "Find edges" which is a 3x3 kernel, also very noise sensitive and relys on only diff with 1.5 pixel region.

-1 -1 -1

-1  8 -1

-1 -1 -1

Filters/Other/Custom allows custom kernels. 

Radial blur technique.

Take multiple images RAW  16bit/channel, align and average to minimize noise .

Copy/Paste, on top image, select Subtract, vary radial blur (try 150 pixels) and level (0.5) to see best result. The result is " second differential" with much lower noise. In addition, merge various exposures, and radial blurs. Sharpen and contrast adjust. Probably best with Sony 7as for low noise. Exposure 1/60 at ISO500 for first test.

"Multiscale Gaussian Normalization" Durkmiller technique

Produces outstanding detail by using many carefully aligned images to reduce noise. Starts with the radial blur.  It is then normalized by dividing local difference by the local std deviation of difference.  Small standard deviation at the zero crossovers enhances the edges. Now merge with multiple radial blurs and multiple exposures. 

https://arxiv.org/pdf/1403.6613.pdf

Pseudocode for MGN 1.

1.Replace spurious negative pixels with zero or local mean/median.                                                Image = Ixy

2. Create Gaussian kernel of width wi . Kernel elements should sum to unity.

3. Convolve image with kernel to create local mean image B ⊗ kw.                                                   Ixy mean = Ave 1-n  Ixy

4. Calculate difference between image and the local mean image, square the difference, and convolve with kernel. Square root the resulting image to give ‘local standard deviation’ image σw (Equation (2)).                                                                                            Std xy =   (Ave 1-n ((Ixy - Ixy mean)^2))^0.5

5. Calculate normalized image Ci by subtracting the local mean image and dividing by the local standard deviation image (Equation (1)). Store the result.

     Edges are created where there is an image and low variation i.e. base of edges                          Ci xy=(Ixy -  Ixy mean )/ Std xy

6. Apply arctan transformation on Ci to give C 0 i .                                                                                   C0 xy = scale Ci xy

7. Repeat 2-6 with the different kernel widths wi .

8. Take mean, or weighted mean if preferred, of the C 0 i to give a weighted mean locally-normalised image.

9. Calculate a global gamma-transformed image C 0 g by applying Equation (4).                              Edge image  Cg xy = scale Ixy

10. Sum the weighted mean locally-normalised image with the global normalized image C 0 g , with appropriate weight h (as Equation (5)).  

                                                                                                                                                                            Final I = Cg xy + C0 xy weighted ave. 

The graphs show a excel simulation of the process on a line image.  Confirming the basic maths of the process. 

Not clear how to get the Stdev in PS. Can get max and min for a pixel area. Max-Min gives a measure of noise, does appear work, needs lower noise.  

Edgeformation.jpg

http://www.zam.fme.vutbr.cz/~druck/Eclipse/TSE2009_Instructions.pdf

It is necessary to take short exposures both after the beginning and before the end of the total eclipse. Exposures longer than 1/30 s should not be taken earlier than ten seconds after the second contact and later than ten seconds before the third contact. For a manually operated camera on a paralactic mount, the following exposure sequence can be used as a good example: 1/500 s (1), 1/250 s (1), 1/125 s (3), 1/30 s (4), 1/8 s (4), 1/2 s (4), 2 s (5), 8 s (5), 4 s (5), 1 s (5), 1/4 s (4), 1/15 s (4), 1/60 s (4), 1/125 s (3), 1/250 s (1), 1/500 s (1). The number in brackets says how many images of the particular exposure time are taken. At an observing site with totality longer than three minutes, it is better to repeat a similar sequence twice of three times. For automated cameras, it is better to change the exposure time for each frame. An example on a good sequence might be: 1/500 s, 1/250 s, 1/125 s, 1/30 s, 1/8 s, 1/2 s, 1/2 s, 2 s, 8 s, 1/60 s, 1/15 s, 1/4 s, 1 s, 4 s Repeat the previous two lines as many times as possible, do not necessarily start from such short exposures each time. The starting exposure may vary from 1/125 s to 1/4 s. Before the end of the eclipse: 1/4 s, 1/15 s, 1/30 s, 1/60 s, 1/125 s, 1/250 s, 1/500 s

Bias images Set the camera to the shortest exposure time the camera enables, the lens must be covered with a cap, ISO setting must be identical with the eclipse images. The number of bias images must be at least four times as high as the total number of eclipse images. Dark-frame images The ISO setting must be identical with the eclipse images, the lens must be covered with a cap. For each image with exposure longer than 1/30 s, there must be at least four dark frames. This means that if four eight-second eclipse images were taken, at least sixteen eight-second dark frames are needed. 5 Flat-field images Do not move the camera much between taking the eclipse images and the flat-field images, so that the dust particles on the chip do not move. (Even if the camera was cleaned in an authorized service, there may still be some dust particles.) Direct the camera to the sky and cover the lens with a diffusor, for example a flimsy paper. ISO setting and aperture value must be identical with the eclipse images. The images must be taken using autoexposure mode with aperture priority with exposure correction −2/3. The number of flat-field images must be at least four times as high as the total number of eclipse images. If the exposure time of the flat-field images is longer than 1/30 s, it is necessary to take dark-frame images for the flat-field images too.

Tse_2013_ed_a.jpg