Digital Darkroom

Good and bad practices (II)

1^st step : normalization

We explained in a few words in another page how to preprocess raw CCD images to substract dark frames and other bias to get frames ready for the post processing. Now, before combining these images they need to be normalized so that they cover an uniform range of brightness. 

The tools are usually automatic algorithms which ensure next functions :

- Display an histogram of the image light curve showing the "brightness" level of each pixel

- Process to stretching to extend the histogram over the full range of data.

In this way the dynamic range of the image will be corrected. The darkest level in the image will be set to the minimum, 0 and the brightness level to the maximum, or 255. This process is similar that using an image processing software or painter and ask it to change the luminosity and contrast of your image.

Upper, adjusting the brightness level of each pixel you can darken the background and increase the subject contrast. At right, in the same way, balancing colors you can give to the sky or the subject a more realistic color, without dominance. Screen dumps are from Cyanogen's MaxIm DL.

A second issue is the color balance of the image. Theoretically the brightest stars must be pure white and the background sky pure black, without red, green or blue dominance. Here also the correction can be set automatically in most imaging programs. In case of algorithm failure, the previous steps must be applied manually, shifting the brightness of each channel. But most of the time the aesthetic effect is improved when the background sky is deep blue. Now our images are prepared and ready for registration and combiantion.

2^d step : registration

The next step is to register all pre-processed images in order to combine their data pixel by pixel without the slightest shift thanks to an alignment tool using two or several points in each image.

The pictures alignment is sometimes a hard-to-match task but a mandatory one if you want to get rigorous results. You will agree to recognize that it will be silly to mix two inverted images of Saturn or images of a galaxy where stars are misaligned...

The principle consists for each image of removing misalignments such as telescope drift, rotation and higher orders differences (including scanner nonlinearity and film unstability using analog peripherals) by positioning peculiar alignment points, typically stars.

This method usually aligns a serie of points in each images with a minimum of 2 points as far apart as possible for a better alignment accuracy. For the ease of the process images are treated two by two and the result image combined together, etc.

Exercise demonstrating the ease of using IP software Using Picture Window Pro 2.5 (now at version 3.5) without reading the manual, but only its features published on Internet, let's enhance the bright areas of a picture of the full Moon in less than one hour. Can you do it? Hints : use a solution for the least exotic in merging an image with a map. An image of the Moon and a lunar map rotated of a few degrees have been registered with Picture Window Pro using a 4-points alignment method. Then the composite at right was created using a Composite transformation (Add) using 75% of simulated exposure for the original and 100% for the mask. At last the resulting image has been slightly post-processed to enhance its sharpness and correct balance (brightness curve). An easy and funny work which result is amazing ! Clic on images to enlarge.

Once the points are in place in each image the program can automatically refine the pixels position. Then using a bilinear interpolation, it creates automatically the composite by shifting the second image on the original. In some cases this alignment algorithm is able to coincide non-square pixels, different camera orientations or different optics with an accuracy down to a fraction of a pixel (the tricks is to zoom the normal view to enlarge the pixel on screen).

Like rotating a single image, an inconvenient of this registration is to soften your resulting image. This side effect can be suppressed by applying a sharpen function with a minimum of strength to restore the original resolution.

3^d step : combination

The basic concept consists of combining images together in respect to the exposure time of each image to simulate a longer exposition. So if you add 4 images, each of them will receive 25% of the weight of the total exposure time.

This step is simple and only the processing tool depends on the nature of your images, B/W or RGB. Indeed, this last often uses a specific function to insure the composite of the three B/W images taken through red, green and blue filters, knowing the blue one requested the longer exposure. 

For B/W images the method consists of using a non alignment points and a blending, adding images according to their time exposure.

Another way to combine images is to use a short exposure image to restore highlights in long exposures. The principle consists of creating a mask that isolates the highlights of the long exposure. This is the famous unsharp mask technique explained in next page, particularly impressing when picturing bright nebulae or the Sun corona during an total eclipse.

If a B/W composite is quite easy to make in stacking many black-and-white frames, an LRGB composite is more subtil as explained in the next link dedicated to CCD's.. 

To increase the contrast of an RGB image, we can take profit of a slight different technique, combining a B/W picture with our RGB's to produce what we call an LRGB image. 

When amateurs speak of LRGB image, theoretically they speak of a processing more sophisticated than simply adding four monochromes frames, 1B/W + 1R + 1G + 1B. In short and theoretically, to increase the signal-to-noise ratio, reduce the turbulence and others artifacts, RGB images should be the combination of several dozen to hundreds of monochromes images. The number is not very important, and in some cases even one RGB is enough when seeing is excellent, but usually, as the seeing is not exceptional, most amateurs prefer to stack several RGB together to reduce the effects of the turbulence in using only one frame (1R+1G+1B). Then this RGB image is combinated with the luminance image. This last gives the contrast to the RGB composite, whithout wich the resulting image looks fine of course but lack of depth; it is not "crisper of details".

The luminance frame should be the combination of a few dozen to hundreds individual B/W frames. This is particulary important when picturing highly featured surfaces like planets (Mars, Jupiter, Saturn and in a lesser extent the crescent of Venus or the one of Mercury).

The final image results then in the combination of all these individual frames. As you can see browsing the image gallery, results can be amazing.

At left a LRGB picture of M104 made by Don Stotz, Mike Ford and Adam Block using a 16" f/10 SCT from NOAO; at center an LRGB picture of M51 made by Tony and Daphne Hallas using a 14" f/8 Cassegrain; at right an LRGB composite of Saturn made by Ed Graphton on Nov 19, 2001 using a C14" at f/68.

At last you can merge several images to circle a vast area of the sky, larger that the field of your scope and create a mosaic. This method requests to extend the input image to enlarge its limits and create a linear gradient mask that runs from black and white across the overlap between the two images. Then you can create a blending mask for the overlay and composite all together, the extended input image, the overlay and the mask.

The basics being defined, in next pages we will see several techniques to improve the aesthetic of astronomical images and to correct their defaults. Unsharp mask, LRGB or mosaic are all terms that apply to image processing techniques that combine multiples photographs to enhance its features. But to get results as exceptional as the ones displayed above or in the gallery, some strict rules must be applied.

To read : Composites by Examples

Back to the Digital Darkroom

Page 1 - 2 -