Taking a single cycle from an image

I have a current waveform image of dimension of 750 * 750. In the current waveform, I have two cycles; before the fault and during the fault. I am only interested in getting the during fault cycle (spike in the current signal). I do not want the figure in gray scale as I need RGB figure. How can I get this cycle only from the images and save in a folder?

[Walter Roberson]

Where should the cycles be considered to start? At the minima? At the zero crossings?

The zero crossing for Phase A before the fault is rouhgly index 125; the zero crossing for Phase B before the fault is roughly 100; the zero crossing for Phase C before the fault is roughly 75. Should each line be cropped separately, or should you crop starting from the first of those (75), or from the last of those (125) ?

[Image Analyst]

What happened to your data? Did you plot it or someone else? Why is your starting point the image instead of the curve values? What do you want as output? Y values in a vector, or a cropped image?

[Md. Nazrul Islam Siddique]

@Walter Roberson, the zero crossing of phase C should be fine. I want a cropped images starting from zero crossing of phase C (75) to the end point (all the lines should be cropped in a single image, not seperately).

[Md. Nazrul Islam Siddique]

@Image Analyst, I want a cropped image.

[Walter Roberson]

Why would the endpoint be the current end of the image? After the fault, the next zero crossings are roughly 170, 190, 210, so shouldn't the endpoint be 210 ?

[Md. Nazrul Islam Siddique]

No, I need the end point for further analysis. Actually I need the full cycle during fault condition.

[Ran Yang]

Sorry for jumping in, but if all OP wants is a cropped figure, can't he just take his current figure and set xlim([start end]) based on whatever he thinks is necessary?

[Md. Nazrul Islam Siddique]

Suppose you have only images about 88,000.

[Ran Yang]

Then going back to what Image Analyst said, do you have the data or not? If you only have images and not the actual data array(s), I don't think anyone can help you.

If you have the data, you can either manually determine the limits for each figure or write a script to do so, and then replot everything.

[Walter Roberson]

If what you have is images rather than figures or data, then I suggest you start by searching the File Exchange for tag:digitize and try out a few of the programs that are created specifically to try to read out coordinates from images.

[DGM]

Do the waveform positions and axes extents vary between images? If not, it should be fairly simple. If they do, it'll be a huge fragile mess, given that this is a difficult JPG.

Answer 1

[DGM]

I'm going to pretend that this won't just explode the moment it's fed a different crusty JPG.

Assuming the following:

• the images all use the same trace colors
• there is a ticklabel for y = 0
• each image is centered on the transient event
• the crop region begins at the last phase C ZC prior to the transient
• the crop region encloses the rest of the trace
• the plot box, ticks, and ticklabels don't need to be kept or reconstructed

inpict = imread('graph.jpg');

% try to get mask of the yellow trace
hsvpict = rgb2hsv(inpict);
lim = [0.103 0.306; 0.066 1; 0 0.985];
lim = permute(lim,[3 2 1]);
tracemask = all(hsvpict>=lim(1,1,:) & hsvpict<=lim(1,2,:),3);

% clean up the mask
tracemask = bwareaopen(tracemask,20);
tracemask = imclose(tracemask,strel('disk',3));
tracemask = bwskel(tracemask);

% blindly try to stitch together all the holes in the mask
% this is based only on blob proximity
% otherwise we have to find a way to distinguish endpoints from many spurs
S = regionprops(tracemask,'pixelidxlist');
sz = size(tracemask);
for k = 1:numel(S)-1
    [y1 x1] = ind2sub(sz(1:2),S(k).PixelIdxList); % all points on the blobs
    [y2 x2] = ind2sub(sz(1:2),S(k+1).PixelIdxList);
    D = (x1.'-x2).^2 + (y1.'-y2).^2; % don't need sqrt
    [minr minc] = find(D == min(D(:))); % find first minimizer
    p1 = [x1(minc) y1(minc)]; % point on blob1
    p2 = [x2(minr) y2(minr)]; % point on blob2
    tracemask = tracemask | brline(sz,[p1; p2]); % draw a line
end

% show it
imshow(tracemask,'border','tight')

%% try to get mask of the plot box, ticks, labels
lim = [0 1; 0 0.5; 0 0.75];
lim = permute(lim,[3 2 1]);
boxmask = all(hsvpict>=lim(1,1,:) & hsvpict<=lim(1,2,:),3);

% try to clean it up
boxmask = imfill(boxmask,'holes'); % fill everything
boxmask = boxmask & ~bwareafilt(boxmask,1); % get rid of plot box
boxmask = imclose(boxmask,ones(21)); % connect each ticklabel as a blob
boxmask(end-100:end,:) = false; % chop off xticklabels
boxmask = bwareafilt(boxmask,1,'smallest'); % get smallest blob (the zero ticklabel)

% show it
%imshow(boxmask,'border','tight')

%% create a mask representing the zero line
S = regionprops(boxmask,'centroid');
yzmask = false(sz);
yzmask(S.Centroid(2),:) = true;

% show it
%imshow(yzmask,'border','tight')

%% find all zero crossings prior to the event
% i assume the event is centered in all plots
zcmask = tracemask & yzmask;
zcmask(:,round(sz(2)/2):end) = false; 

% get the position of the last ZC in this window
S = regionprops(zcmask,'centroid');
startcol = round(S(end).Centroid(1));

% show it
%imshow(zcmask,'border','tight')
%hold on; plot(S(end).Centroid(1),S(end).Centroid(2),'o'); hold off

%% extract the ROI
alltracemask = hsvpict(:,:,2)>0.04; % can't always assume phase C defines extents
tracerows = any(alltracemask,2);
startrow = find(tracerows,1,'first');
endrow = find(tracerows,1,'last');
endcol = find(any(alltracemask,1),1,'last');
outpict = inpict(startrow:endrow,startcol:endcol,:);

% show it
imshow(outpict,'border','tight')

I disabled the display of the masks, since they are all single-pixel features and the figures are all downscaled with nearest-neighbor interpolation. Run the code yourself if you want to see anything.

Of course, nothing can be done about the fact that part of the plot is actually missing due to the legend box, which is now unreadable.

[Md. Nazrul Islam Siddique]

Thank you. I will try it.