Change color of specific bits Imagesc(), Image()

Hi,

I have a binary matrix, where my algorithm runs through to find inconsistencies and flip the respective bits as it goess row by row. I want to be able to show those specific bit changes in a different color to visually demonstrate the process. Is it possible?

Edit - I may have missed out a key point. I use imagesc() to show the entire array. In that figure, I want the bit change to be shown. Currently I am playing with casting a bit sized rectangle on the flip coordinates

[Rishi Balasubramanian]

For those bit flips, i flip it using the ~ operator, so I will be having the coordinates of the specific bit flips to begin with.

Answer 1

[Walter Roberson]

Yes. See https://www.mathworks.com/matlabcentral/fileexchange/24093-cprintf-display-formatted-colored-text-in-command-window

[Rishi Balasubramanian]

I may have missed out a key point. I use imagesc() to show the entire array. In that figure, I want the bit change to be shown. Currently I am playing with casting a bit sized rectangle on the flip coordinates

Answer 2

[Image Analyst]

I see you've added tags about images. So, see my attached demo about viewing bit planes. If you flipped a bit, you could have another logical matrix that keeps track of what pixels got their bits flipped, and then use imoverlay to display the flipped bits in color over the original.

rgbImage = imoverlay(originalMatrix, matrixOfFlippedBits, 'r');
imshow(rgbImage);
title('Flipped bits in red');

[Rishi Balasubramanian]

My process tends to collect upto 10^9 and above rows with a minimum col size of 126, so storing more might hurt my speed.

[Image Analyst]

Yes, fitting 126 billion numbers (one terabyte) in an array in memory would be a challenge. How are you doing it? Are you using memmapfile()?

[Walter Roberson]

10^9 rows on a display that probably is not more than 2000 available pixels requires that half a million rows be displayed per pixel. Could you talk more about how you plan to encode the information for half a million rows in one pixel?

If you are lucky enough to have an 8K display (rare but possible) that is still 125000 rows per pixel.

[Rishi Balasubramanian]

Well, I am not looking to simulate all 10^9 * 126 in the graph, It is a long process and If I could simulate a few, yeah

[Walter Roberson]

Okay, suppose your simulation gets down to 9 data values per available screen pixels, and suppose for the moment that you have 2000 screen pixels -- so (9 * 2000) x 126 input values. Not much data, really -- less than 18 megabytes.

So you have 9 values to summarize per pixel, and each one might have been flipped or might not have been flipped.

Each of the 9 values being summarized into a single pixel can have at least 2 states. That requires at least 2^9 different possible values for the output pixel. Which is 512 distinguishable values.

Are your users going to be able to reliably figure out which of 512 different brightnesses or color combinations at particular pixel is? Probably not.

[Rishi Balasubramanian]

Well, for the time being, I am using the rectangle function to create an oversized rectangle (3 bits width) with the location for bit flips and colored it red and increased the thickness to make it look like a really ugly red blop on the approximate error locations.

That's what I could muster. This is not for the users, but for my own reference for debugging. Just wanted to see approximately where the flips are happening instead of stopping the code and getting into the array in the workspace every now and then.

Now that I have you here, I have another question pposted regarding communications toolbox - https://www.mathworks.com/matlabcentral/answers/1690395-bch-encoder-error-locations?s_tid=srchtitle

Think you could also help with this?

[Walter Roberson]

You have indicated that you have up to 10^9 rows. You are trying to represent those rows visually. I have pointed out that your display has only a small fraction of that resolution.

You could assign each bit to be a "summary" of a range of rows. For example you might decide:

• first color: no errors in this range, values were majority 0
• second color: no errors in this range, values were majority 1
• third color: errors have occurred in this range, the majority of them flipped 0 to 1
• fourth color: errors have occurred in this range, the majority of them flipped 1 to 0

You do not need to keep track of all of the bits in a block: you could keep track of the number of 0's and 1's and 0-1 errors and 1-0 errors.

You could even animate, keeping statistics over a range of time; decrement by totals counters by the stored numbers from N states ago to "reverse" the effect of that previous state.