I want to recognize only the handwriting numbers (blue color). I attached my code, it can only recognize the printed numbers. I used Mnist database to recognize them. How to recognize only the handwriting numbers?

clear all

clc

load('MnistConv.mat');

I = imread('testnumber3.jpg');

% flipImage=flip(I,2);

% rotate=imrotate(flipImage,180);

image(I)

% Convert to grayscale image

Igray = double(rgb2gray(I));

% Convert to binary image

Ibw1 = im2bw(Igray,graythresh(Igray));

Ibw=(256-Igray)/256;

% Ibw = im2bw(Igray,graythresh(Igray));

% Edge detection

Iedge = edge(Ibw,'Roberts');

% Morphology

% * *Image Dilation*

se = strel('square',2);

Iedge2 = imdilate(Iedge, se);

% * *Image Filling*

Ifill= imfill(Iedge2,'holes');

[Ilabel num] = bwlabel(Ifill);

disp(num);

Iprops = regionprops(Ilabel);

Ibox = [Iprops.BoundingBox];

Ibox = reshape(Ibox,[4 num]);

% Plot the Object Location

hold on;

for cnt = 1:num

kotak=rectangle('position',Ibox(:,cnt),'edgecolor','r');

crop=imcrop(Ibw,Ibox(:,cnt));

% Resizing

imgLGE = imresize(crop, 5, 'bicubic');

X = abs(imresize(imgLGE, [28 28]));

for i=1:28

for j=1:28

if X(i,j)<0.

X(i,j)=0;

elseif X(i,j)>0.5

X(i,j)=X(i,j)+0;

end

end

end

k=1;

x = X(:, :, k); % Input, 28x28

y1 = Conv(x, W1); % Convolution, 20x20x20

y2 = ReLU(y1); %

y3 = Pool(y2); % Pool, 10x10x20

y4 = reshape(y3, [], 1); % 2000

v5 = W5*y4; % ReLU, 360

y5 = ReLU(v5); %

v = Wo*y5; % Softmax, 10

y(:,cnt) = Softmax(v); %

end

[max,result]=max(y);

Answer 1

[Prabhan Purwar]

Hey,

Assumption: This solution is justified only for cases having handwritten images in a different color to that of computer-generated and background.

Regionprops() function in MATLAB works well for BW images with well-defined boundaries as illustrated in its algorithm.

• https://www.mathworks.com/help/images/ref/regionprops.html(regionprops)

In the present scenario input to the function is as follows:

In order to get defined boundaries, make use of Color Thresholder app (using colorThresholder command) with spectrum settings as shown

• · https://www.mathworks.com/help/images/ref/colorthresholder-app.html(Color Thresholder)

Output after giving the following image as input is shown below

Once we segregate proper boxsets, by thresholding them based on box area, they can be further given as input to the neural network for recognition.

Although, more advanced techniques can be employed for this task, as linked below

• https://www.mathworks.com/help/vision/examples/recognize-text-using-optical-character-recognition-ocr.html(Recognize Text Using Optical Character Recognition (OCR))
• https://www.mathworks.com/help/deeplearning/examples/character-recognition.html(Character Recognition)

[DONY DARMAWAN PUTRA]

Thanks Prof Prabhan Purwar, your solution is amazing. It is very helpfull to solve my problem.