look into colour calibration and/or colour casts with your software. For me I use PixInsight (got neb and ST but never get far with them personally) so I register (align), integrate (stack) my dslr images this gives me a 64bit depth integration.fit file to start with. I suppose I should look into debayering as step 1 before registration. But since I dont PI has a tool that removes the small green cast since my bayer pattern has two greens to each red/blue pixel the cast isnt obvious at first but it gets boosted up as the image goes through my workflow.I also do a colour calibration step
Your Carina has a red cast though (maybe white balance settings?) but you might find something to help there, try stretching your histogram first by separate R, G & B channels, I expect your R channel differs greatly. When you get the three humps aligned it may be its more natural state and you can then adjust RGB as one again. if this makes sense.
Linear etc, well bayered is the raw image from your camera, dslr usually use a bayer matrix of RGBR, so try with a single raw sub. As Rick said Linear is the right choice from what youve done. I dont understand why its called "linear" but I see it in terms of ratios, in a single sub a pixel in the middle of a star is probably overexposed to pure white, a pixel in the background should be pure black, and perhaps the arms of a galaxy a 50% grey (go with me on the rough numbers here please). So the colour distance between the black, grey and white pixels are say 1:1:1. Now you align and stack a second frame, the ratio stays 1:1:1, align and stack 1,000 frames and it stays at 1:1:1. So the result doesnt "look" any better than when you started with a single frame.
Its in its linear state still. What you have gained that you cant yet see is depth.
Typical displays are only 8bit depth (8 bits per colour channel per pixel, so 8x R+8xG+8xB gives us our 24bit display spec. now 8 bits is 256 values
so going back to my example the white pixel starts at 256 value (well actually 256, 256, 256 for colour) which is as much as a computer display shows, align and stack second frame the value stacks to 512 but gets translated back down for display to 256, think of it as value divided by stacked frames. Likewise the black pixel will have a value of 0 and the gray 128. Now the captured frames contain signal which we want (so star centre of white is 256,256,256) and noise which we dont. The noise is added into the signal and stored as our image. Every so often a bit of noise will sit on our white pixel changing it slightly to off white (say 250, 253, 256). But mostly the white pixel should always be white. So going back to stacking giving us depth where value divided by stacked frames all the values of this whit pixel should eventual average out close to the true expected value of 256,256,256, likewise black to 0,0,0 and the grey to 128,128,128.
This depth gives you a huge flexibility to stretch values before the noise values start reimposing themselves into your image since its still there in the signal. And faint signals can be revealed too , and false faint signal from stacked noise clumps get pushed into the depths. So it increases the accuracy of the values in the image, but as I said at the start it stays in a linear display format so wont look any better on its own. The depth it provides lets you be more adventurous with the processing so you can stretch it further to a "nice" display image without the noise becoming dominant again.
Confused now?
The terminology drove me crazy when I started astrophotography processing. Image registration? I was looking for the online database where I would have to register and wonder how much it'd cost. Similiar/same terms crop up if you do any medical imaging too btw.
Again as Rick said "linear means it hasn't been stretched yet". I always keep my first linear image as its the starting point and contains everything in my subs set. Later on with new knowledge and software I can load in the same linear image and reprocess to get improvements.
). So much to learn...
