Automatically normalize a range of data into specific values.

The system I am working with collects brightness data from a LED that uses three brightness levels to encode binary data along with a clock. The data range is 0 - 255. My intention is extract binary values (along with the clock value) from this range automatically; is it possible to do so?

Example raw brightness data:

[20,   253,   253,   108,   109,   254,   253,   254,    17,    19,   253,   253]

Example working code:

% Normalize raw luminous intensity values into a binary and clock
% representation. (zeros, ones, and twos)
% Note: Equation parameters found through observation.

disp('Normalize data into zeros, ones, twos')

for i = 1:length(raw_data)

         if raw_data(i) < 50 
          normalized_data(i) = 0;
         end

         if (raw_data(i) > 50) && (raw_data(i) < 200)
           normalized_data(i) = 1; 
         end

          if raw_data(i) > 200 
           normalized_data(i) = 2;
         end
  end   

disp (normalized_data)

Output:

[0, 2, 2, 1, 1, 2, 2, 2, 0, 0, 2, 2]

Is there a way to decide the equation parameters programmatically? Currently, I am bringing up a graph of the raw_data and through observation I am deciding the range where the zeros, ones, and twos should fall.

edit

I should note that lows, mediums, and highs of the raw data vary depending on external conditions and leaving constants as parameters may cause problems in a finalized system.

Thank you.

Answer 1

[Vivek Selvam]

Yes, it is possible using hist for finding the low, medium and high ranges. I have used your code as a function to calculate the normalized data based on the ranges. It can be optimized using vectoring and logical indexing.

function normalized_data = autoNormalize(raw_data)
% Note: Equation parameters found through histogram binning.
    brightnessLevels        = 3;
       % get number of elements and center of each level
    [numElements,centers]   = hist(raw_data,brightnessLevels);
       % plot a bar graph - to view what happened
    bar(centers,numElements)
       % compute level edges
    width                   = diff(centers);
    edges                   = centers(1:end-1) + width/2;
       % call normalizeMe to classify
    normalized_data         = normalizeMe(raw_data,edges);
end

function normalized_data = normalizeMe(raw_data,edges)
% Normalize raw luminous intensity values into a binary and clock
% representation. (zeros, ones, and twos)
    disp('Normalize data into zeros, ones, twos')
    for i = 1:length(raw_data)
           % modified from < to <= to take care of edge cases 
        if raw_data(i) <= edges(1) 
        normalized_data(i) = 0;
        end
           % similar modification
        if (raw_data(i) > edges(1)) && (raw_data(i) <= edges(2))
        normalized_data(i) = 1; 
        end
        if raw_data(i) > edges(2)
        normalized_data(i) = 2;
        end
    end   

      disp(normalized_data)
  end

[Reuben]

Thank you very much, the code works excellent! I am going to take the time and look deeper to why the code works. =)

Answer 2

[Image Analyst]

Try imquantize() in the Image Processing Toolbox - it's meant for this purpose.

[Reuben]

Thank you, I will look into this function.