40m QIL Cryo_Lab CTN SUS_Lab TCS_Lab OMC_Lab CRIME_Lab FEA ENG_Labs OptContFac Mariner WBEEShop
  TCS elog  Not logged in ELOG logo
Entry  Sat Jun 19 17:31:46 2010, James K, Misc, Hartmann sensor, SURF Log -- Day 3, Initial Image Analysis 7x
    Reply  Tue Mar 1 10:47:18 2011, Aidan, Misc, Hartmann sensor, Electron to Counts conversion efficiency 
Message ID: 53     Entry time: Sat Jun 19 17:31:46 2010     Reply to this: 116
Author: James K 
Type: Misc 
Category: Hartmann sensor 
Subject: SURF Log -- Day 3, Initial Image Analysis 
For Friday, June 18:
(note that I haven't been working on this stuff all of Saturday or anything, despite posting it now. It was getting late on Friday evening so I opted to just type it up now, instead)

(all matlab files referenced can be found in /EDTpdv/JKmatlab unless otherwise noted)

I finally got Xming up and running on my laptop and had Dr. Brooks edit the permissions of the controls account, so now I can fully access the Hartmann computer remotely (run MATLAB, interact with the framegrabber programs, etc.). I was able to successfully adjust camera settings and take images using 'take', saving them as .raw files. I figured out how to import these .raws into MATLAB using fopen and display them as grayscale images using the Imshow command. I then wrote a program (readimgs.m, as attached) which takes inputs a base filename and number of images (n), then automatically loads the first 'n' .raw files located in /EDTpdv/JKimg/ with the inputted base file name, formatting them properly and saving them as a 1024x1024x(n) matrix.

After trying out the test pattern of the camera, I set the camera into normal operating mode. I took 200 images of the HWS illuminated by the OLED, using the following camera settings:

Temperature data from the camera was, unfortunately, not taken, though I now know how to take it.
The first of these 200 images is shown below:

As a test exercise in MATLAB and also to analyze the stability of the HWS output, I wrote a series of functions to allow me to find and plot the means and standard deviations of the intensity of each pixel over a series of images. First, knowing that I would need it in following programs in order to use the plot functions on the data, I wrote "ar2vec.m" (as attached), which simply inputs an array and concatenates all of the columns into a single column vector.

Then, I wrote "stdvsmean.m" (as attached), which inputs a 3D array (such as the 1024x1024x(n) array of n image files), which first calculates the standard deviation and mean of this array along the 3rd dimension (leaving, for example, two 1024x1024 arrays, which give the mean and standard deviation of each pixel over the (n) images). It then uses ar2vec to create two column vectors, representing the mean and standard deviation of each pixel. It then plots a scatterplot of the standard deviation of each pixel vs. its mean intensity (with logarithmic axes), along with histograms of the mean intensities and standard deviation of intensities (with logarithmic y-axes).

"imgdevdat.m" (as attached) is simply a master function which combines the previous functions to input image files, format them, analyze them statistically and create plots.

Running this function for the first 20 images gave the following output:

(data from 20 images, over all 1024x1024 pixels)

Note that the background level is not subtracted out in this function, which is apparent from the plots. The logarithmic scatter plot looks pretty linear, as expected, but there are interesting features arising between the intensities of ~120 to ~130 (the obvious spike upward of standard deviation, followed immediately by a large dip downward).

MATLAB gets pretty bogged down trying to plot over a million data points at a time, to the point where it's very difficult to do anything with the plots. I therefore wrote the function "minimgstat.m" (as attached), which is very similar to imgdevdat.m except that before doing the analysis and plotting, it reduces the size of the image array to the upper-left NxN square (where N is an additional argument of the function).

Using this function, I did the same analysis of the upper-left 200x200 pixels over all 200 images:

(data from 200 images, over the upper-left 200x200 pixels)

The intensities of the pixels don't go as high this time because the upper portion of the images are dimmer than much of the rest of the image (as is apparent from looking at the image itself, and as I demonstrate further a little bit later on). Do note the change in axis scaling resulting from this when comparing the image. We do, however, see the same behavior in the ~120-128 intensity level region (more pronounced in this plot because of the change in axis scaling).

I was interested in looking at which pixels constituted this band, so I wrote a function "imgbandfind.m" (as attached), which inputs a 2D array and a minimum and maximum range value, goes through the image array pixel-by-pixel, determines which pixels are within the range, and then constructs an RGB image which displays pixels within the range as red and images outside the range as black.

I inputted the first image in the series into this function along with the range of 120-129, and got the following:

(pixels in intensity range of 120-129 in first image)

So the pixels in this range appear to be the pixels on the outskirts of each wavefront dot near the vertical center of the image. The outer circles of the dots on the lower and upper portions of the image do not appear, perhaps because the top of the image is dimmer and the bottom of the image is brighter, and thus these outskirt pixels would then have lower and higher values, respectively. I plan to investigate this and why it happens (what causes this 'flickering' and if it is a problem at all) further.

The fact that the background levels are lower nearer to the upper portion of the image is demonstrated in the next image, which shows all intensity levels less than 70:
(pixels in intensity range of 0-70 in first image)

So the background levels appear the be nonuniform across the CCD, as are the intensities of each dot. Again, I plan to investigate this further. (could it be something to do with stray light hitting the CCD nonuniformly, maybe? I haven't thought through all the possibilities)
The OLED has been turned off, so my next immediate step will be to investigate the background levels further by analyzing the images when not illuminated by the OLED.
In other news: today I also attended the third Intro to LIGO lecture, a talk on Artificial Neural Networks and their applications to automated classification of stellar spectra, and the 40m Journal Club on the birth rates of neutron stars (though I didn't think to learn how to access the wiki until a few hours right before, and then didn't actually read the paper. I fully intend to read the paper for next week before the meeting).
Attachment 2: ar2vec.m  202 Bytes  Uploaded Mon Jun 21 08:12:07 2010  | Hide | Hide all
function V = ar2vec(A)
%AR2VEC V=ar2vec(A)
%concenates the columns of 2D array A into a single column vector V

sz = size(A);

while i<(n+1)
... 7 more lines ...
Attachment 3: readimgs.m  971 Bytes  Uploaded Mon Jun 21 08:13:03 2010  | Hide | Hide all
function arr = readimgs(imn,n)
%- A function to load a series of .raw files outputted by 'take'
%and stored in /opt/EDTpdv/JKimg/
%  Inputs: 'basefilename' is a string input (for example, for series of
%   images "testpat####.raw" input 'testpat'). "n" is the number of images,
%   so for testpat0000-testpat0004 input n=5

... 32 more lines ...
Attachment 4: stdvsmean.m  973 Bytes  Uploaded Mon Jun 21 08:13:25 2010  | Hide | Hide all
function M = stdvsmean(A)
%STDVSMEAN takes a 3D array of image data and computes
%stdev vs. mean for each pixel

%find means/st devs between each image
astd = std(double(A),0,3);
armn = mean(double(A),3);

%convert into column vectors of pixel-by-pixel data
... 33 more lines ...
Attachment 5: imgdevdat.m  312 Bytes  Uploaded Mon Jun 21 08:13:45 2010  | Hide | Hide all
function imgdevdat(basefilename,imgnum)
%IMGDEVDAT Inputs base file name and number of images stored as .raw files
%in ../EDTpdv/JKimg/, automatically imports as 1024x1024x(n) matrix, finds
%the mean and standard deviation of each pixel in each image and plots

Attachment 6: minimgstat.m  345 Bytes  Uploaded Mon Jun 21 08:14:08 2010  | Hide | Hide all
function imgdevdat(basefilename,imgnum,size)
%IMGDEVDAT Inputs base file name and number of images stored as .raw files
%in ../EDTpdv/JKimg/, automatically imports as (size)x(size)x(n) matrix, finds
%the mean and standard deviation of each pixel in each image and plots
Attachment 7: imgbandfind.m  702 Bytes  Uploaded Mon Jun 21 08:35:57 2010  | Hide | Hide all
function [HILT] = imgbandfind(img,minb,maxb)
%IMGBANDFIND inputs an image array and minimum and maximum value,
% then finds all values of the array within that range, then plots with
%values in range highlighted in red against a black background

... 20 more lines ...
ELOG V3.1.3-