Determine the axis range of the graph image

I'm working on a small project on digitizing an image of a graph. I have a graph image as input and output as coordinate value (x,y) pairs of the graph in that image. Now I'm having a hard time determining the range of values of the axes. I want to use OCR to recognize the labels along the axes. But I don't know how to implement it. Can someone help me?? Here is an example of a graph image. I want to determine the min and max values of the x-axis (-20 and 20) and the y-axis (-1 and 6)

Answer 1

[yanqi liu]

sir，please check the follow code to get some information

for more information,you can read the book 《计算机视觉与深度学习实战》，Thank you！

clc; clear all; close all;
im = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/777828/image.jpeg');
bw = im2bw(im);
bw = ~bw;
bw2 = bwareaopen(bw, 1200);
bw(bw2) = 0;
bw = logical(bw);
bw3 = imclose(bw, strel('line', round(0.05*size(bw,1)), 90));
bw3 = imclose(bw3, strel('line', round(0.05*size(bw,2)), 0));
rs = sum(bw3, 2); cs = sum(bw3,1);
[~,ind1] = max(rs); [~,ind2] = max(cs);
y = 1:size(bw,1); x = 1:size(bw,2);
pt1 = [x(:) repmat(ind1,length(x), 1)];
pt2 = [repmat(ind2,length(y),1) y(:)];
bw3 = bwselect(bw3, [pt1(:,1); pt2(:,1)], [pt1(:,2); pt2(:,2)]);
bw(~bw3) = 0;
bw4 = imclose(bw, strel('line', round(0.05*size(bw,2)), 0));
bw41 = bwselect(bw4, [pt1(:,1)], [pt1(:,2)]);
bw42 = bwselect(bw4, [pt2(:,1)], [pt2(:,2)]);
bw42(bw41) = 0;
bw41 = logical(bw41); bw42 = logical(bw42);
[r1,c1] = find(bw41); [r2,c2] = find(bw42);
[~,ind1] = min(c1); [~,ind2] = max(c1);
[~,ind3] = min(r2); [~,ind4] = max(r2);
bw5 = bwselect(bw4, [c1(ind1) c1(ind2) c2(ind3) c2(ind4)], [r1(ind1) r1(ind2) r2(ind3) r2(ind4)]);
[L,num] = bwlabel(bw5);
stats = regionprops(L);
wh = 5;
bws = [];
figure; imshow(im);
for i = 1 : num
    bwi = bw;
    bwi(L~=i) = 0;
    recti = stats(i).BoundingBox;
    hold on; rectangle('Position', recti, 'EdgeColor', 'r', 'LineWidth', 2, 'LineStyle', '-');
    recti = [recti(1:2)-wh recti(3:4)+2*wh];
    bws{end+1} = imcrop(bwi, round(recti));
    % use ocr or cnn
end

figure;
montage(bws, 'Size', [2 2], 'BackgroundColor', 'w', 'BorderSize', [3 3])

[Trong Link]

Thank you very much! Your code is very helpful to me but if the labels of the axes are small size (as shown below) it is not possible to get the correct position of them. Is there any way I can improve this. And if the axes are logarithmic, how can the ocr() function recognize it as a logarithmic value?

The x-axis is a logarithmic scale:

The y-axis is a logarithmic scale:

[yanqi liu]

yes，sir，if we can get the figure，may be use the handle to get data

of course，if just use image，the image may be use imresize to make it bigger，and use some interactive to get the effect value