MATLAB: How to adjust box size of blob analysis

Question

i have a picture of digits. i have croped that digits using blobanalysis then i resize that croped image to 20×20 and after place that 20×20 image into 28×28 image so that i can pedict that digit. but there is a problem when there is 1 in image and the box form around 1 is very small so when i resize that croped 1 it becom disorted. how can i fix this

clc;
clear;
BackgroundImage = imread('back.png');
object = imread('11.png');
subplot(3,3,1);
imshow(object);
title('Orignal Image');
ga = rgb2gray(object);
BW = im2bw(ga);
subplot(3,3,2);
imshow(BW)
title('convert im2bw');
gb = rgb2gray(BackgroundImage);
foregroundDetector = vision.ForegroundDetector('InitialVariance',(30/255)^2);
foreground = step(foregroundDetector, gb);
foreground1 = step(foregroundDetector, ga);
BlobAnalysis = vision.BlobAnalysis('MinimumBlobArea',100,'MaximumBlobArea',50000);
[area,centroid,bbox] = step(BlobAnalysis,foreground1);
Ishape = insertShape(object,'rectangle',bbox,'Color', 'green','Linewidth',6);
subplot(3,3,3);
imshow(Ishape);
no_of_digits = size(bbox,1);
pred = zeros(1,no_of_digits);
for k = 1:no_of_digits
    im_k = imcrop(ga,bbox(k,:));
    im_k = imresize(im_k,[20,20]);
    white = 255 * ones(28, 28, 'uint8');
    white(5:24,5:24) = im_k;
    final_image = white;
    subplot(3,3,k+3)
    imshow(final_image)
    pred(k) = pred_number1(final_image);
    imshow(final_image)
end

Thanks 🙂

Answer 1

The shape of the bounding box will always give the smallest rectangular area in which the blob is contained, but you could change your code a bit to make it so that the blob is not distorted. What you do instead of resizing it to a specific size is adapting the scale in imresize. Followed by inserting that image in the white padded image. A rough implementation in your code by replacing a few lines would look like this:

for k = 1:no_of_digits
    im_k = imcrop(ga,bbox(k,:));
    % resize based on scale:
    maxSize = max(size(im_k));
    im_k = imresize(im_k,20/maxSize);
    
    white = 255 * ones(28, 28, 'uint8');
    % assign the image inside the white one:
    % get the size of the resized image
    sizeK = size(im_k);
    % determine the indices to place it in the middle of the padded image
    iStart = floor((28-sizeK(1))/2)+1;
    iEnd = iStart+sizeK(1)-1;
    jStart = floor((28-sizeK(2))/2)+1;
    jEnd = jStart+sizeK(2)-1;
    white(iStart:iEnd,jStart:jEnd) = im_k;
    
    final_image = white;
    subplot(3,3,k+3)
    imshow(final_image)
    pred(k) = pred_number1(final_image);
    imshow(final_image)
end