Assamese Handwritten Digits

Citation Author(s):
Prarthana
Dutta
NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR
Naresh Babu
Muppalaneni
NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR
Submitted by:
Prarthana Dutta
Last updated:
Wed, 11/06/2019 - 04:14
DOI:
10.21227/n441-bk63
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

This dataset comes up as a benchmark dataset for machines to automatically recognizing the handwritten assamese digists (numerals) by extracting useful features by analyzing the structure. The Assamese language comprises of a total of 10 digits from 0 to 9. We have collected a total of 516 handwritten digits from 52 native assamese people irrespective of their age (12-86 years), gender, educational background etc. The digits are captured in .jpeg format using a paint mobile application developed by us which automatically saves the images in the internal storage of the mobile. The images are of variable size and are further modified as per requirment. It contains a total of 516 images, among them 387 images (75%) are taken in the training set and 129 images (25%) in the testing set. 

The Convolutional Neural Networks are extensively used for the purpose of classification as it can learn millions of internals features automatically and perform classification.

Instructions: 

In this dataset there are 10 different classes from 0 to 9 of the Assamese numerals. All of the 516 images are of varying dimensions which are further modified by cropping, making all the images of identical dimension to fit the character.

The labels are attached in the word document. 

                          

০ - 0

১ - 1

২ - 2

৩ - 3

৪ - 4

৫ - 5

৬ - 6

৭ - 7

৮ - 8

৯ - 9