Collecting Photons

IC 1848

Megrez 80 Super Apo with 0.8 TV field flattener: 51x8' in two different nights, iso 800.

Hyperbolic asin stretch, masked processing in PixInsight. Final touches in Photoshop CS. See the following for processing details.

Back Home

Drowning in a sea of noise

A single subframe shows that the nebula is actually kind of bright, since it can be well observed in a single frame.

The summed stack before gradient subtraction

Two nights and 79 sub exposures later we have a complete stack. The stack is shown here at about twice the stretching of the final image to show the lack of uniformity of the background. Typically, my raw images suffer from two problems: 1) a strong greenish gradient, which is typically brighter on the side closer to the horizon. This is mostly due to light pollution; 2) a red glow in the corners, due to the amplifier glow of the camera which is not completely removed by the dark subtraction.

Masking the nebula.

I prepare a binary mask in Photoshop to improve the gradient computation performed by Iris. In practice, the black areas are NOT used during the gradient computation.

After the gradient subtraction.

This is the resulting image after gradient subtraction. In this specific case I run the process twice, at first with a low order polynomia to take away most of the gradient, that has very low spatial frequency. A second pass with a higher order polynomia reduces the sharpest changes, especially at the corners of the image. The image is strongly stretched to show off the correction. With this technique I have corrected the background of two different datasets.

Two datasets are better than one.

The total data set is so large that a large fraction of stars are totally saturated in the complete stack, even if they are not so in the original raw frames. This is due to the worst limitation of Iris: its dynamic range limited to 15 bits. I computed two different stacks: a) σ/k sum of the entire dataset, this image has lots of saturated pixels belonging to the brightest star but it has the best possible S/N in the low lights; b) σ/k sum of the 16 best frames. This sum was performed in Iris with the normalisation flag set on. In practice the sum is stacked at 32 bits depth and at the end the result is normalized to avoid saturation. Of course this determines the loss of the low order bits, but this is not relevant since this stack will only be used to obtain luminance and colour information of the brightest stars. The two stacks have been corrected for gradient, as detailed above, and white balanced by neutralising the background and applying an RGB correction computed for my modified Canon 350D (1.15, 1.00, 1.24).

Correcting for chromatic aberration

Before any non linear transformation of the data, it is necessary to complete the pre-processing. At this stage we can correct the small, but still present, chromatic residual of my apochromatic refractor. The chromatism can be detected at the corners of the image, and it appears as a consequence of a slight difference in focal length at the blue and red extremes. This can be corrected by separating the images in the RGB components and by a registration with an affine transformation, that align for translation, rotation and linear change of scale. Move the mouse over the image to see the before/after images (scaled at twice the original size for clarity).

Non linear stretching in Iris

The two stacks (complete stack and normalised stack) are stretched with the inverse hyperbolic sine function of Iris. This stretching allows for an optimal separation of the chroma values, since it enhances the differences between the RGB channels. The complete stack received a moderate stretching (asinh 0.002), while the normalized stack was pushed a little further (asinh 0.003) to improve the separation of the different spectral classes. These are details of the frame center shown at full scale. Rolling over with the mouse shows the two stacks. The complete stack shows a much brighter nebula at the expenses of saturating the brightest stars (about one fourth of the total). The normalized stack shows little nebula but not a single star in the field is saturated.

Adjusting and blending the layers

There are many different ways of combining the two data set, both in Photoshop CS and in PixInsight. In this case I opted for an easy way: in brief, the two images are stacked as two separate layers in CS, with the complete stack below. Working with the magic wand on the top layer, I selected the brightest areas that resulted to be stars only. After carefully controlling the extent of the selected area I feathered it (feather 3, in this case), inverted it and cancelled. In this way the nebula and the dimmer stars became visible underneath. The operation is completed by lifting the low lights of the upper stack with the curves command in order to improve the blending between the two layers. Rolling over the detail with the mouse shows the blended stack alternating with the original stack. Notice that the brighter stars are not saturated and a bit smaller.

Fine adjustment of levels and noise control in PixInsight

The final adjustment on the histograms are operated with PixInsight which is much more accurate than photoshop. At this stage I also apply some noise reduction masking the high and medium lights) and a moderate erosion on the high lights to compact the star images. Finally I apply a slight adjustment of the curves. Roll over the image to see the image before/after the PixInsight processing.

Send anything that comes to mind to Gimmi Ratto gimmi@in.cnr.it