shivang8 / digit-recognition Goto Github PK

Hello shivang8 , thats good coding. I am trying to figure out how to use new data as a input .

first problem is that i want to use 28x28 .bmp picture whichis created with paint. so i create function which takes the image converts data to logical array in format of 1x784 . here it is

`  function [s] = row_and_col_values_of_img(test_picture)
  s=zeros(1,784) ;
 I =(test_picture)  ;
          for j=1:28  
                 for i=1:28  
            s(1, i+ 28*(j-1) )=I(j,i); 
                 end
              end
 end ` 

since i used ".bmp" format with one colour all my values are logical.

then i used weights and biases to calculate output ..

nn = load('data_epoc200.mat');
test_folder = 'C:\Users\enesk\Desktop\w-s\nn-matlab\50piksel\28piksel'; % folder with big images
files_1 = dir(fullfile(test_folder));

Picture_set_1=imread(char(join( (test_folder)+"\"+files_1(3).name ) ) );
s = row_and_col_values_of_img(Picture_set_1);

weight1= nn.weight1;
weight2=nn.weight2;
bias1= nn.bias1;
bias2=nn.bias2;
neth = s * weight1;
neth = neth + bias1;
outh = sigm(neth);
% Output Layer
 neto = outh * weight2;
 neto = neto + bias2;
 outo = sigm(neto);

but the output values are not matching with the image label ....

my picture is 5.

but the output array gives me something like this

0,00890543561907159 0,00137493916567926 0,000313196127740974 0,00203675665040161 2,50346713054253e-06 0,000663910983342224 0,000115502557563390 0,000172873298425822 0,956766243598225 0,000439554708199857

closest value to 1 is 9th values so it says my icture is 9 ....