STAR REMOVAL & ANALYSIS TOOL (OpenCV)

Concept:

The idea of this tool is to remove all stars and the surrounding halo of any deep sky picture. The result should be just an image with the deep object, and another with the star mask.

It is not only interesting to visualize the object without stars; it could be also interesting to perform some image adjustments only on the deep sky object (e.g. color adjustments, sharpening, etc) without the influence of overexposed stars. Later, adjusted deep sky image can be merged back with star mask image.

After all stars are identified, it could be possible to perform a guiding quality analysis studing the geometry of the star images. It could also be possible to define a vectorized image that shows guiding quality and also optical deviations in the whole image; e.g. at the edges the stars are more oval towards the edge than the stars in the center. This is due to the optical lenses of the system.

It could also be possible to determine the background gradient, as clearly seen on the original image (center is brighter than the edges). It could be possible to flat this gradient...somehow ;-)

Status:

Only some tests with the slightly modified floodfill algorithm were done; just to identify the bright star regions.

Process:

The stars are identified and marked with a red line. I used the floodfill algorithm which I implemented for the planetary stacking software to mark the contour of the stars.

There is a threshold value that determines the brightness of the stars to be recognized (it´s not a real magnitude [mag] value!).

Original Image with many stars:

Stars are recognized and marked with red line:

I'm still thinking about several ways to fill the extracted star regions. The easiest way could be filling the region with the average color of the region. But if the star is bigger, then this monochromatic color without noise would also be visible.

Another way could be to add artificial noise, calculating de deviation if it from the background.

A better way could be finding similarity of the ROI in the whole image, and just fill the extracted star doing copy paste of the most similar region.

Another way cold be analyzing the surrounding of the extracted star, and fill the empty region in iterations following the surrounding pixels texture, pixel by pixel.

Problems:

- find & implement the best filling algorithm. Could be not that simple if some stars are inside the deep sky object or at nebulas border regions, e.g. Messier 27, the Dumbbell Nebula or Messier 31, the Andromeda Galaxy.

- identify and remove not only the bright stars, but also the colored halo that surrounds it

- leave nebulas region and other deep sky objects as original as possible

- many other problems I still not faced now...