Text recognition from a dot-matrix display

I have a video from which I extract image frames at a specified time interval. It is a video of reading from an oxygen online analyzer. I need to recognize the text in each extracted image using MATLAB and append that to a text file, instead of manually looking into each image and noting down the reading in a text file. I have tried to binarize the image and do OCR on the region containing the text alone (by getting the position of the region of interest using getPosition and drawRectangle). I have also used "seven-segment" as language model. Despite these, the recognized text is either empty character array or an array with irrelevant numbers. Can you suggest some techniques like pre-processing for better OCR of such images?

vid = VideoReader('D:\ChemE\MTech\MTP\PurityData\Closed_Pressurization_Vacuum\0.5LPM\30_c1.mp4');
numFrames = vid.NumberOfFrames;
n=numFrames;
fps = 30; %frames per second
fileID = fopen('D:\ChemE\MTech\MTP\PurityData\Closed_Pressurization_Vacuum\0.5LPM\30_c1.txt','w'); %open text file for writing
titl = "Time(s)     Purity(%)"; % for the text file which will be appended with the text from OCR
fprintf(fileID,"%s\n",titl);
frames = read (vid,1); %read the first frame of the video to identify ROI 
figure;imshow(frames);
rect = drawrectangle; % to select the region of interest (ROI)
roi = rect.Position;
figure(close);
for i = 1:fps:n
    frames = read(vid,i); %read the ith frame (at interval of 1 second)
    % set a file name for images
    fname = ['D:\ChemE\MTech\MTP\PurityData\Closed_Pressurization_Vacuum\0.5LPM\30_c1\30_c1_' int2str(i),'.jpg'];
    imwrite(frames,fname); %writes the frame to an image
    I = imread(fname); % read the image
    bI = imbinarize(rgb2gray(I)); %binarize the image after converting it to a grayscale image
    ocrResults = ocr(bI,roi,Language = "seven-segment"); %recognize text
    val  = ocrResults.Text; %store the recognized text
    conf = ocrResults.WordConfidences;
    fprintf("The purity observed is ""%s"" with confidence %.4f \n", val, conf);
    fprintf(fileID,'%.1f\t %s\n',(i-1)/30,val); %write the text to file
end

Answer 1

[Gojo]

Hi Sai,

It seems that you are extracting images from a video and performing OCR by selecting a region of interest to recognize the text. Since you have already tried binarizing the image, try preprocessing the image before performing OCR.

I have saved the image that you have attached in the question and tried to preprocess it. Have a look at the following steps:

1. Import the image and convert it to grayscale:

img = imread('ocr_img.jpeg'); %this is the image present in the ques
grayI = rgb2gray(img); % Convert the image to grayscale 

2. Following that you could apply a filter to reduce noise. For example, try a median filter:

filteredI = medfilt2(grayI); % Noise reduction 

For more information on filters, you could try referring to the following documentation: https://www.mathworks.com/help/images/noise-removal.html#:~:text=Get-,Kmedian,-%3D%20medfilt2(J)%3B%0Aimshowpair

3. Adjust the contrast and sharpen the image:

contrastAdjustedI = imadjust(filteredI); % Contrast adjustment 
sharpenedI = imsharpen(contrastAdjustedI); % Sharpening 

For more information about “imadjust” and “imsharpen” you can refer to the following documentation links:

https://www.mathworks.com/help/images/ref/imadjust.html

https://www.mathworks.com/help/images/ref/imsharpen.html

4. As a final preprocessing step, binarize and dilate the image to fill gaps. Prior to dilation, complement the image. Have a look at the following code snippet:

bI = imcomplement(sharpenedI); 
% Create a structuring element (adjust the size as needed) 
se = strel('disk', 6);  

% Dilate the image to fill in gaps 
dilatedImage = imdilate(bI, se); 
filledImage = imcomplement(dilatedImage); 

For more details about “imdilate” please find the documentation here: https://www.mathworks.com/help/images/ref/imdilate.html

5. For retrieving the text through OCR, you can follow the same process of creating a region of interest as shown in the following code snippet:

rect = drawrectangle; % to select the region of interest (ROI) 
roi = rect.Position; 
textStruct = ocr(filledImage,roi,Language = "seven-segment"); 
text = textStruct.Text; 
disp(text); 

Using the above steps, I was able to retrieve the numerical value present in the image.

I hope this helps!