Classification of MNIST database (MATLAB Code)

Description

One of the popular database in image processing is MNIST. The user often cannot read this database correctly and cannot access to the images in this database. The files in this database are :

t10k-images.idx3-ubyte

t10k-labels.idx1-ubyte

train-images.idx3-ubyte

train-labels.idx1-ubyte

Some images of this database are shown in the following image:

We get ready a MATLAB code for you to read train and test images.

Then feature extraction has been done on input images.

Then a classifier is trained by training data.

Then a classifier is applied to test data.

Confusion matrix of output is :

C =

    83     0     0     0     0     0     1     0     0     1

     0   125     0     0     1     0     0     0     0     0

     1     0   101     2     1     0     2     9     0     0

     1     0     3    96     0     3     0     0     3     1

     1     0     0     0   102     0     1     0     0     6

     0     0     0     2     0    82     0     1     2     0

     1     0     0     0     0     0    85     0     1     0

     0     1     2     4     0     1     0    88     0     2

     0     0     1     0     0     0     3     0    77     8

     0     0     0     0     1     0     0     2     2    9

Performance of test data is 92.90 % .

You can use other classifiers in this code.