Problems with optical character recognition

I am implementing an automatic graph image digitizer. I'm trying to determine the labels on the axes, I use the built-in function ocr() to determine them. It can initially determine the exact values if the x and y axes are linear scales. But if the axes are logarithmic scale then the ocr() function cannot determine the exact logarithmic values on each axis. Is there any way I can get those logarithmic values?

Here is the images that I cropped from my input graph image:

How can the logarithmic values be determined on those images?

[Image Analyst]

What does your ocr() give? Like 107, 106, 105, etc. If so you'll just have to recognize that the third number is the power of ten.

[Trong Link]

I can do so if I know the axes are logarithmic scale. I'm assuming my digitizer doesn't know what scale the axes are, linear or logarithmic.

Answer 1

[yanqi liu]

yes，sir，if we can get the figure or .fig file，may be use figure handle can get the property value

but，if just image，it is an digits ocr problem，the logarithmic data type may be use the height to classify between normal digits、logarithmic digits

when get the class，use image segment to get the number list，and make the right number or the small height number as exponential number

here is an example

clc; clear all; close all;
img = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/830925/image.png');
img = rgb2gray(img);
bw = imbinarize(img,'adaptive','ForegroundPolarity','dark','Sensitivity',0.4);
bw2 = imdilate(imclose(~bw, strel('square', 7)), strel('square', 7));
stats = regionprops(bw2);
rect1 = stats(1).BoundingBox;
bwi = imcrop(bw, round(rect1));
bwi = imresize(bwi, 40/size(bwi,1), 'bilinear');
statsi = regionprops(~bwi);
ocrResults=ocr(bwi,'CharacterSet','0123456789')

ocrResults = 
  ocrText with properties:

                      Text: '107↵↵'
    CharacterBoundingBoxes: [5×4 double]
      CharacterConfidences: [5×1 single]
                     Words: {'107'}
         WordBoundingBoxes: [8 8 40 24]
           WordConfidences: 0.6906

str = strtrim(ocrResults.Text)

str = '107'

rectsi = cat(1, statsi.BoundingBox);
rectsi = sortrows(rectsi, 1);
if rectsi(end, 4)/rectsi(1, 4) < 0.8
    % the right number height 
    str = [str(1:end-1) '^' str(end)];
end
str

str = '10^7'

figure; imshow(img, [])
hold on; rectangle('position', rect1, 'EdgeColor', 'g', 'LineWidth', 2)
text(rect1(1), rect1(2), str, 'color', 'r')