Aumentar la señal de una imagen----Enlarge an image signal

[edulopez]

No se si será correcto lo que propongo o si ya se hace.
Es un procedimiento muy simple pero que a primera vista ofrece unos resultados interesantes.
partiendo en este caso de una imagen  de la nebulosa del caballo, resultado de la suma de 12 tomas de 5 minutos con QHY8 y fs78 a f8 2,54 arcseg/pix.

Bueno, lo que he hecho ha sido hacer cuatro copias de la imagen original, a una de ellas le he eliminado dos lineas de pixeles superiores, a la siguiente dos lines de píxeles inferiores,a otra dos colúmnas de píxeles en el lado derecho y a la última dós columnas en el lado izquierdo, con lo que me quedo con una imagern  "entera" y cuatro recortadas. El siguiente paso es hacer una suma con Pixelmath, he probado a ha cer la suma de varias maneras y creo que la que mejor resultado da es esta....      (imegen original+recortada1+recortada2+recortada3+recortada4)/5..Creo que es mejor hacer esto antes de procesar, pero aun no lo he comprobado
bueno a probar...... He probado además a eliminar en vez de dos píxeles 1 y 3 pero a primera vista con dos ofrece mejor resultado, será cuestión de ir probando.gracias por leerme


Do not know if what I propose is correct or if it is done.
It is a very simple procedure that at first glance but offers some interesting results.
parties in this case an image of the nebula of the horse, the result of the sum of 12 shots in 5 minutes with QHY8 and fs78 to f8 2.54 arcsec / pix.

Well, what I did was make four copies of the original image, one of them I've deleted two lines of pixels above, the following two lines of pixels lower in another two columns of pixels on the right side and last two columns on the left side, so I stay with a imagern "whole" and four cutouts. The next step is to do with PixelMath sum, I have tried to sum cer in several ways and I think that gives best result is this .... (IMEG recortada1 + original + recortada2 + recortada3 + recortada4) / 5 ..
good to try ...... I also tried to remove two pixels instead of 1 and 3 but at first sight with two offers better result will be a matter of going to read me. thanks

[Carlos Milovic]

Hola Edu

Para ser más precisos, no haz aumentado la señal. Se ha disminuido el ruido
La operación que propones es idéntica a una convolución pasa bajos, esto es, un filtro de desenfoque. Un filtro gaussiano, una convolución del tipo promedio (o estadística, como la mediana) te darán resultados muy similares, sino idénticos.
El aumento en el contraste del resultado es nada más que un resultado de reescalar. Probablemente no estabas usando todo el rango dinámico al empezar.


Quick english answer: this operation is almost the same as a low pass convolution. Gaussian, average or even median filters should give the same (or very similar) results.

[edulopez]

Gracias por aclararme las dudas.....

++++Please clear my doubts ...

[Niall Saunders]

Hi all,

When I first read the message this morning (and didn't have time to reply) I thought much the same - it is almost (but not quite) the implementation of a low-pass convolution - but with only the 'sides' (in fact the outside edges) of the 3x3 kernel being used.

In fact, my initial thought would be to compare it directly with the RestorationFilter process, specifically using a Gaussian Point Spread Function (PSF), and more specifically with StdDev set to 0.75 and Shape set to 5.0 (everything else can be lef at Default settings). This gives the necessary 3x3 'kernel', with very 'dark corners' being equivalent to no 'corner' processing being applied in the PixelMath example, and with something like '50%' of the 'side edges' being used.

In any case Edu, you might be interested in 'playing' with this process, and seeing how it can affect your images. The background of this type of process is explained in detail in the "Handbook of Astronomical Imaging", by Richard Berry and James Burnell.

Cheers,

[Carlos Milovic]

Hi Niall

The RestorationFilter performs quite the opposite. It is a deconvolution, not a convolution You are trying to "revert" (or invert) the effect of a low pass filter, modelated through the PSF. Edu's method is a plain convolution, with a "strange" factor due the size scalations performed internally by PixelMath, but nothing that cannot be reproduced with a custom made kernell. Such a process has not been implemented yet, even when it was incorporated in our very first PI Std Anyway, it should be here anytime soon (Juan's standards).

[Juan Conejero]

Just a few points. When PixelMath applies an image whose dimensions are different from the dimensions of the target image, then it interpolates its pixels to match the target image's geometry. This is done automatically. This means that you can for example add two images whose dimensions are 200x200 and 100x100; resampling of the data is done automatically on the fly, as necessary to match the sizes.

For this reason what you've applied, as noted in the answers above, is actually a low-pass filtering process. Pixel interpolation always works as a low-pass filter in PixInsight.

As for the signal, it cannot be improved through image processing. In fact, it is just the opposite: each time you modify your data through image processing, the recorded signal is degraded in some way. Hopefully, we can modify the data in a way that the noise is reduced more than the signal, the uncertainty in the data is hence reduced, and the data is transformed into information. This is the main purpose of image processing.

[Niall Saunders]

Hi Carlos,

I must admit that, as I was typing my reply, and since I pressed the <Post> button, I had started to think the same as you. In other words I was 100% "WRONG" 

However, life is a process of learning - and this is a PI Process that I have not really played with at all, so maybe now is the time to have a closer look,

Cheers,