log_temp

STEP 1
	set up imsaves
Step 2
	Look at stars
		Add check in dark process to flag peak vals
			What does the 90th percentile mask look like?
Step 3
	Look at radial sections


% starfield analysis: In which we look at the subset of the image with
% distinct points and devise a few tests.
% We can use a chiral integration kernel on the direction of the gradient.
% The kernel returns 1 (topological filter) when it encloses a peak
% I can then use this kernel to find all isolated peaks. For ones nearby -
% start with the known and suspected stars - we can use them as a common
% ground truth to crack the PSF, noise correction, etc.
%     Then test to see if we can determine stars from hot pixels!
%             Expect correlation between channels, not just one colour?
% I guess, you know, a gaussian filter would do fine to find them... surely
    % ah but you gotta pick teh small ones

Todo
	Small dev
		Radial intensity plotting (/contour intensity plotting)
		Correlation fn: Stars vs hot pixels
		Remove zeroing, in fact apply NO noise correction and see what top edge looks like
		Use regression over the noise offset to find the correction for noise & exposure time
			you KNOW the underlying intensity has to be the same, so you can find the 
	Look into Baileys Beads...
		Red channel, indiv. imgs, optimize correction via corrfun
	What about looking at correlation betweem imgs *inside* the disk of the sun?

	Bundle out function that
		Gradient discontinuities...
		Takes im gradients & FFT
		Plots:
			Raw 		num_samples
			I grad 		I grad dir  	
			FFT 		Phase						
			r grad 		theta grad
		Plot radial (and angular? sections through centre)
	Apply to:
		Darkfield
		Sum img at various stages
		save to:
			stage/img/diagnostic
		All raw images to find intrinsic features?
	Cluster of bright points, below disk and bottom left


Observations
	Intensity gradient 'wobbles' in interior I_grad - noise offsetting and/or saturated threshold cutoff errs?
		Saturation seems less likely the problem, so focus on noise & alignment 1st
	Red splotches in the sky... ~ sensor noise scale or no?
	The gradient colours (black/white) seem to be shorter-wavelengths *in the direction of* the differential vector??
	What's up with the intensities?
		Several (angular and radial) are VVV flat in the exposure area - I guess that's good, you can use it as a mask??
		Coincides with one of the few grey ares in the circumferential filter
	What are the amplitudes of these noise features? They are hard to spot in the colour image
	Noise patterns differ in each case
		- What about setting 
	Short-wavelength noise
		Period ~few pixels?
	
	Medium-wavelength,
		~dozens of px?
		Noise correction: of zeroing out, include all and then subtract from the mean
			Should *increase* scale of blob functions in gradients
				Could use cluster size as a cost function
					Maximize wrt Noise cutoff
	Both present in the radial & angular gradients
		Correlation function between different gradient types should give length scales

	flat spots in gradient img: fewer samples? 0 values?
		Can use some integral as error fn
			Could it be related to alignment?
				Draw box around imgs
					Output num_samples
	Circumferential gradient
		Thin bright lines! Saturated areas? Prob not. Consistent brightness though, v strong!
			Black sections - negative gradients? Need to see in colour/bicolor?
	Cartesian gradient
		Same lengths scales present - but pattern structure slightly different?	


Consider:
	Before noise correction, take correlations of gradients?
	Ah - theory for the edges -
		They are where the inner image was below noise floor and set to zero in that area
			So would need to know