error in corr2 function in matlab

I have created templates of number and symbols and want to match the vehicle number plate with template using corr2 function. While running the code got the following error

Error using corr2
Expected input number 1, A, to be two-dimensional.

MAtlab code for template creation

%CREATE TEMPLATES
one=imread('1.png'); 
one=imresize(one,[42 24]);
two=imread('2.png');
two=imresize(two,[42 24]);
three=imread('3.png');
three=imresize(three,[42 24]);
four=imread('4.png');
four=imresize(four,[42 24]);
five=imread('5.png'); 
five=imresize(five,[42 24]);
zero=imread('0.png');
zero=imresize(zero,[42 24]);
sign=imread('sign.png');
sign=imresize(sign,[42 24]);
number=[one two three four  five zero ];
character=[number sign];
NewTemplates1=mat2cell(character,42,[24 24 24 24 24 24 24], 3);
save ('NewTemplates1','NewTemplates1')
clear all

matlab code for reading the letter

load NewTemplates1 % Loads the templates of characters in the memory.
snap=imread('untitled.png');
snap=imresize(snap,[42 24]); % Resize the input image so it can be compared with the template's images.
comp=[ ];
for n=1:length(NewTemplates1)
    sem=corr2(NewTemplates1{1,n},snap); % Correlation the input image with every image in the template for best matching.
    comp=[comp sem]; % Record the value of correlation for each template's character.
end
vd=find(comp==max(comp)); % Find the index which correspond to the highest matched character.
%*-*-*-*-*-*-*-*-*-*-*-*-*-
% Accodrding to the index assign to 'letter'.
% Alphabets listings.

if vd==1 
    letter='1';
elseif vd==2 
    letter='2';
elseif vd==3
    letter='3';
elseif vd==4 
    letter='4';
elseif vd==5
    letter='5';
elseif vd==6
    letter='0';
elseif vd==7
    letter='sign';
end

For kind information: The size of individual component in template is 42x24x3 uint8 and the input image also same dimension. Please provide the suggestion to do further.

[Mansi Sharma]

can you please provide me the correct code for this ? I'm getting same problem.

[wiqar Ahmad]

can you please share me the correct code?

Will be Thankfull

[Walter Roberson]

As discussed below: use rgb2gray() before doing the corr2.

Answer 1

[Walter Roberson]

You are asking corr2 to operate on a 42 x 24 x 3 array. That is a 3d array which is not something that corr2 can use. Your array is 3d because you are carrying through rgb. You need to do each plane independently, or you need to switch to greyscale, or you need to reshape so that you are working in 2d.

[SHOBA MOHAN]

The code works fine after doing rgb2gray conversion. Thanks Walter

Answer 2

[John BG]

Hi Shoba Mohan

this is John BG <mailto:[e-mail] [e-mail]>

Since the actual size of the input images, both the references and the unknown input to correlate with the references, I have omitted the imresize lines.

With that, one way to have your code running is as follows:

clear all;clc

one=imread('1.png');one=one(:,:,1);
% one=imresize(one,[42 24]);
two=imread('2.png');two=two(:,:,1);
% two=imresize(two,[42 24]);
three=imread('3.png');three=three(:,:,1);
% three=imresize(three,[42 24]);
four=imread('4.png');four=four(:,:,1);
% four=imresize(four,[42 24]);
five=imread('5.png'); five=five(:,:,1);
% five=imresize(five,[42 24]);
zero=imread('0.png');zero=zero(:,:,1);
% zero=imresize(zero,[42 24]);
sign=imread('sign.png');sign=sign(:,:,1);
% sign=imresize(sign,[42 24]);
number=[one two three four  five zero ];
character=[number sign];

NewTemplates1=zeros([size(one) 7]);
% mat2cell(character,42,[24 24 24 24 24 24 24], 3);
NewTemplates1(:,:,1)=one;
NewTemplates1(:,:,2)=two;
NewTemplates1(:,:,3)=three;
NewTemplates1(:,:,4)=four;
NewTemplates1(:,:,5)=five;
NewTemplates1(:,:,6)=zero;
NewTemplates1(:,:,7)=sign;

save ('NewTemplates1','NewTemplates1')

clear all
load NewTemplates1 % Loads the templates of characters in the memory.
snap=imread('untitled.png');snap=snap(:,:,1);
% snap=imresize(snap,[42 24]); % Resize the input image so it can be compared with the template's images.
comp=[ ];
[sz1 sz2 sz3]=size(NewTemplates1);
for n=1:sz3
    sem=corr2(NewTemplates1(:,:,n),snap); % Correlation the input image with every image in the template for best matching.
    comp=[comp sem]; % Record the value of correlation for each template's character.
end
[vd,col,q]=find(comp==max(comp)); % Find the index which correspond to the highest matched character.
%*-*-*-*-*-*-*-*-*-*-*-*-*-
% Accodrding to the index assign to 'letter'.
% Alphabets listings.
if vd==1 
    letter='1';
elseif vd==2 
    letter='2';
elseif vd==3
    letter='3';
elseif vd==4 
    letter='4';
elseif vd==5
    letter='5';
elseif vd==6
    letter='0';
elseif vd==7
    letter='sign';
end

Now the code doesn't return error but the correlation doesn't catch some of the samples I have used.

Wouldn't it be reasonable to consider already working MATLAB application for car plate recognition, like:

https://uk.mathworks.com/matlabcentral/fileexchange/44088-car-license-plate-reader

.

thanks in advance for time and attention

John BG

<mailto:[e-mail] [e-mail]>

[SHOBA MOHAN]

Thank you very much for the answers and providing the links to be refereed John B G. The code given in the links works well for extracting number,however, My request is to extract the person with disability sign along with number from car number plate . Can you please suggest how to do this ?

[Walter Roberson]

Extracting just the red channel from each png is not very useful in locations such as Ontario Canada, or California USA, which mostly use blue license plate numbers and letters for non-commercial vehicles (in Ontario, diplomatic plates are white font and commercial vehicles are black font.)

[Walter Roberson]

In my country, current personal license plate texts are:

• British Columbia: blue
• Alberta: red
• Saskatchewan: green
• Manitoba: blue
• Ontario: blue
• Quebec: blue
• Nova Scotia: blue
• New Brunswick: red
• Prince Edward Island: light brown as of 2013 but green for decades before that so most would be green
• Newfoundland and Labrador: blue as of 2007, but red for the 14 years before that so a mix would be expected
• Yukon: black
• Northwest Territories: blue
• Nunavut: black

So your version would work with Alberta; New Brunswick; some older Newfoundland and Labrador that would not be rare; Yukon; Nunavut; and a small minority of Prince Edward Island. In terms of population: somewhere on the order of 15% to 20%, I estimate.

[Walter Roberson]

The rules about license plates on the front of cars are provincial in Canada, so some provinces require them and some do not.

[John BG]

Thanks again for this other clarification :)