Recognition of Handwritten Assamese Characters

Saurabh Sutradhar
Research Scholar, Discipline of Computer Science,
Hiranya Chandra Bhuyan School of Science and Technology, KKHSOU

Dr. Ridip Dev Choudhury

Associate Professor, Discipline of Computer Science
Hiranya Chandra Bhuyan School of Science and Technology, KKHSOU

Abstract. Handwriting recognition has earned a fair amount of attention in the field
of pattern recognition. A lot of work has been done so far for the efficient recognition of
handwritten characters. So far, various techniques have been put forward for
handwritten character recognition. However, not much work has been done yet for the
recognition of Assamese handwritten characters because of their complex structure and
limited usage. The Assamese language spoken in some of the northeastern parts of
India is lagging in this field. This paper aims to make human Assamese handwriting
machine-readable. Successful recognition will enable machines to read forms, postal
addresses, signboards written in As- samese. It will also lead to advancements in other
areas such as translations of handwritten Assamese scripts to other languages as well as
enable a text to speech system, thus making the Assamese language more
acknowledgeable. In this pa- per, we have attempted to recognize handwritten
Assamese characters using Con- volution Neural Network (CNN) by collecting our
dataset. We have judged the performance of our system by the performance metrics:
Precision, Recall, and F1 score.

Keywords: Assamese Characters, CNN, Classification, Dataset.

1 Introduction
Handwritten character recognition is a technique of detecting, segmenting, and identi- fying
characters from images. The main objective of handwritten character recognition
2

is to replicate the human reading capabilities so that the computer can understand, read
and work as humans do with text. It has been one of the most interesting and laborious
research areas in the field of image processing and pattern recognition lately. Several
research works have been focusing on new methods that would reduce the processing
time without compromising high recognition accuracy.

A lot of work for handwritten character recognition for languages like European,
Chinese, and Arabic has been done. Genetic algorithm and Artificial Neural Network
(ANN) were used as a classifier to recognize English digits and alphabets in [6]. Deep-
Learning was also used to recognize English digits and alphabets in [10]. Even for do-
mestic Indian languages like Hindi, Bangla [4], Oriya [2], Malayalam [5], Tamil [11],
etc. a lot of research has been done till now. Indic scripts like Oriya, Telugu, Bangla,
Roman have already been explored extensively [2,4].

The Assamese language has not been much explored due to its less usage. The As-
samese language is a part of the Assamese-Bengali Script prominently used in the
North-Eastern region of India. The Assamese script consists of 11 vowels and 40 con-
sonants [15]. It also has 122 conjunct characters [15]. Recognition of Assamese Char-
acters is a difficult task because of the complex features of the characters like convo-
luted edges (ক্ষ, শ, ঋ, etc.), similarity in appearance ( অ, আ, য, য়, presence of
loops ( ভ, ঞ, etc.), etc. Unlike English script, the Assamese script does not have the
etc.),
concept of capitalization of first characters. Also, different people have different hand-
writing styles therefore it becomes a very laborious task of determining the characters.

Nowadays, CNNs have been effectively applied to pattern recognition, image clas-
sification, and forecast studies to name a few. For the classification of Assamese digits
and characters, a few versions of CNN have been used so far [13,14]. This research
area of recognizing Assamese characters has not been explored much. The potential of
this field is immense considering the popularity of the language in its native state.

We have used the performance metrics of precision, recall and F1 score, for our
calculation. The formulas of the respective performance metrics are given below:

Precision = Tp/ (Tp + Fp)

Recall = Tp/ (Tp+ Fn)
F1 = 2 X [(P X R) / (P + R)]
Where, Tp = Number of true positives
Fp = Number of false positives
Fn = Number of true positives (Tp) + Number of false positives (Fp)

Additionally, this paper is divided into five sections. The first is the Related Work sec-
tion, where we have given a detailed survey in this area. Next, is the Methodology
section where we have explained the workflow of our system. We have also discussed
our dataset and explained each of the steps we have thoroughly used in our system. It
 3

is followed by the Results and Discussion section which focuses on the results obtained
in our research. In the end, we have the conclusion and future scope section where we
have underlined the future breadth of this research area.

2 Related Work

Printed and handwritten Assamese digits were compared in [1]. For the feature ex-
traction and classification, part feed-forward neural networks and a tree-based classifier
were used. Digits were scanned from the document after pre-processed images were
cropped with a bounding box to obtain individual digits. Features were extracted from
each bounded area of the image. The grayscale documents were converted into a binary
image and the background noises were removed with the help of linear filtering, medial
filtering, and adaptive filtering. After that skew detection and correction were per-
formed, following line, word, and character segmentation.

Word-level script identification for six handwritten Indic scripts (Bangla, Devana-
gari, Gurmukhi, Malayalam, Oriya Telugu, and Roman) was proposed in [2]. This pa-
per proposed the elliptical and approximation approach to design features. The original
images are in grayscale and Otsu’s Global thresholding method is used to convert them
into binary images, removing the noisy pixels from the binarized image using Gaussian
Filter. For the classification part, the feature sets have been applied to 7 different clas-
sifiers namely- Naïve Bayes, Bayes Net, MLP, SVM, Random Forest, Bagging, and
Multi-Class Classifiers. The multi-layer perceptron (MLP) achieved the highest accu-
racy of 94.35%.

A computational model for Handwritten Assamese text was proposed in [3] to per-
form the preprocessing, text segmentation, and then extraction of different features
from individual characters. To segment, a document image into various parts, a global
projection profile approach of a word was used to identify the upper, lower, and middle
areas of a word. A combination of diagonal features using zoning concept and texture
features via GLCM (Grey Level Co-occurrence Matrix) were computed for extracting
various features of individual characters. The mean and standard median filters (SMF)
were used to clean noises from an input image. Random transform-based techniques
were used for skew detection and correction. An artificial neural network as a backend
was used to perform classification and recognition tasks.

A modified ResNet-18 architecture (a convolution neural network architecture) was

proposed in [4] to recognize Bangla handwritten characters. This paper considered three
main challenges: Recognizing convoluted edges to distinguish between the repetitions
of the same pattern in different characters and different handwritten patterns for the
same characters. To have wider input for the generalized performance of the network,
input images are preprocessed through the removal of noise with the median filter, edge
thickening filter, and the image is resized to a square shape with appropriate paddings
4

by default. The dataset used for this paper is the two recently introduced datasets-
Bangla Lekha-Isolated dataset and CMATERdb dataset.

A hybrid approach for recognizing Malayalam handwritten characters was proposed

in [5]. It takes into account both the dependent and independent features of the lan-
guage. MATLAB is used to implement the proposed OCR system. The proposed
method of OCR is a hybrid approach for feature extraction combining both structural
and statistical features. In this paper, the curve features are extracted using the water
reservoir principle and a decision tree classifier is used for classification.

An Artificial Neural Network (ANN) based approach was used to segment hand-
written text in Assamese [7]. After the feature extraction part, the input was fed to an
ANN model, and the similarity measure was found. A text-to-speech synthesizer was
used in [8] to facilitate English text reading. The text was manually typed in the screen
and the system was made using MATLAB. Handwritten Sindhi numerals were simi-
larly recognized in [9] using K-NN and SVM. They evaluated their system using the
correlation coefficient. The Sindhi numeral 0 achieved an accuracy of 100% whereas
the numeral 3 achieved 63%. Then for Tamil script, features like character height,
width, slope, etc. were considered in [11]. Zernike moments were used for feature ex-
traction and then fed to a backpropagation model.

Similarly, for the classification of Assamese handwritten digits, vowels, and conso-
nants a zoning feature was used in [12]. A feed-forward neural network model was used
with a sigmoid function at every neuron to calculate the output. For digits, vowels, and
consonants a recognition accuracy of 70.6%, 69.62%, and 71.23% was achieved re-
spectively. In [14] Assamese handwritten digits were recognized by using CNN (Digi-
Net model). This paper used six alternative Convolution and Max-Pooling layers which
were later followed by a Fully Connected layer and a Softmax classifier. CNN was
further used in [13] by using LeNet-5, ResNet-50, Inception V3 and DenseNet-201
model.

3 Methodology and Dataset

The proposed system consists of the following phases:

1. Data Collection
2. Digitization and Normalization
3. Preprocessing
4. Convolution neural network / Classification

The block diagram is given in Figure1

 5

Fig. 1. Workflow diagram of our system

3.1 Data collection

The collection of datasets is the first and most necessary part of the work. We went
to a local school to ask small children to write the Assamese characters in different font
sizes, for our research work. People from 7 years and above were asked to fill in the
boxes of 0.75 cm2 ,1.0 cm2 , and 1.25 cm2. We wanted to include diversity and variation
in our dataset as much as possible.

We have included the snapshot of a page of the dataset in fig. 2.

6

Fig. 2. A sample page of our collected dataset

3.2 Digitalization and normalization

After the collection of datasets, the image of every character was taken by our phone
cameras. The images were then manually cropped. For every character, 225 images
were collected. So, our total dataset consists of 9,225 images. Each image of the
dataset is fixed to a size of 50 x 50.

3.3 Preprocessing

The images were converted into a grayscale image. The color of the images was then
inverted. For noise removal, we used Gaussian blur and Otsu’s thresholding. Gaussian
blur was used for smoothening and Otsu’s thresholding was used for filtering in our
system.

Fig. 3. Before Preprocessing

 7

Fig. 4. After Preprocessing

3.4 CNN / Classification

We divided our dataset into two parts in an 80:20 ratio, namely training and testing.
The training dataset consists of 7,380 images and the testing dataset consists of 1,845
images. After dividing the datasets, the training images are fed to Convolution Neural
Network.
A CNN pipeline model usually is made of a feature learning phase and a classifica-
tion phase. The feature learning phase consists of the Convolution and Pooling layers.
It learns the high-level features of our pre-processed input image. The Convolution
layer performs linear operations in the input image for feature extraction. Then we have
used the relu and softmax functions as the activation function to ensure non-linearity in
our network. Then for the pooling layer, we have used the max-pooling function to
reduce the dimensionality of the network and detect the strong features. The convolu-
tion and pooling operations are performed on every pixel of the input image inde-
pendently. The output of the feature learning phase is a flattened column vector.

Since we have to classify 41 characters (consonants), our output layer consists of 41

nodes. Then for the classification phase, a feed-forward network must be established.
It is achieved by mapping the flattened column vector to the output layer with the aid
of a fully connected layer. This fully connected layer learns the logic behind the feature
learning phase and performs the classification of Assamese characters. We have used
five layers in our CNN network, the dropout and dense layers being alternatives. We
went for categorical_crossentropy and RMSprop() for the loss function as the opti-
mizer.
In the testing phase, the testing dataset (1845 images were from the test folder) was
compared with the training dataset. We tested for 41 Assamese alphabets

4 RESULTS AND DISCUSSIONS

We evaluated the performance of the classification by the metrics accuracy, recall, pre-
cision, and F1 score.
8

We went for a batch size of 128 with epochs of 15. The result of our system has been
shown in Table 2.

Table 1. Performance measure of our system

ACCURACY PRECISION RECALL F1 SCORE
70.02% 68.5% 69.23% 70.71%

5 Conclusion and future scope

We have discussed the progress in the field of Assamese handwritten character recog-
nition in this paper. In our research work, we have achieved an accuracy of 70.02%,
precision of 68.5 %, recall of 69.23%and F1 score of 70.71% by using a convolution
neural network. The accuracy can be further improved by increasing the size of the
dataset, adding more pre-processing steps, or by using a hybrid model.
In the future, the Assamese handwritten character recognition system can further be
extended by adding a voice feature in the post-processing part. The recognition part
could also be used to identify conjunct characters shortly. After successful recognition
of conjunct characters, words and sentences can be targeted next.

References
1. Medhi, K., Kalita, S.: Assamese Digit Recognition with Feed Forward Neural Network. In-
ternational Journal of Computer Applications. Volume 109 - No. 1 (January 2015).
2. Singh, P., Doermann, D.: Word-level Script Identification for Handwritten Indic Scripts.
13th International Conference on Document Analysis and Recognition (ICDAR). IEEE,
2015
3. Bania, R.: Handwritten Assamese Character Recognition using Texture and Diagonal Ori-
entation features with Artificial Neural Network. International Journal of Applied Engineer-
ing Research 13.10 (2018): 7797-7805.
4. Alif, M., Ahmed, S., Hasan, M.: Isolated Bangla Handwritten Character Recognition with
Convolutional Neural Network. 20th International Conference of Computer and Information
Technology (ICCIT). IEEE, 2017.
5. Sujala, K., James, A.Saravanan, C.: A Hybrid Approach for Feature Extraction in Malaya-
lam Handwritten Character Recognition.Second International Conference on Electrical,
Computer and Communication Technologies (ICECCT). IEEE, 2017
6. Agarwal, M., Kaushik, B.: Text Recognition from Image using Artificial Neural Network
and Genetic Algorithm. IEEE, 2015
7. Bhattacharya, K., Sarma, K.: ANN-based Innovative Segmentation Method for Handwritten
Text in Assamese. IJCSI International Journal of Computer Science Issues, Vol. 5, 2009
8. Gopinath, J., Aravind, S., Chandran, P., Saranya, S.: Text to Speech Conversion System
using OCR. International Journal of Emerging Technology and Advanced Engineering, Vol-
ume 5, Issue 1, January 2015
9. Sanjrani, A., Baber, J., Bakhtyar, M., Noor, W., Khalid, M.: Handwritten Optical Character
Recognition System for Sindhi Numerals. IEEE 2015
 9

10. Vaidya, R., Trivedi, D., Satra, S.: Handwritten Character Recognition Using Deep-Learning.
2nd International Conference on Inventive Communication Technologies (ICICCT). 2018
11. Wahi, A., Sundaramurthy, S., Poovizhi, P.: Handwritten Tamil Character Recognition. Fifth
International Conference on Advanced Computing (ICoAC). 2013
12. Medhi, K., Kalita, S.: Assamese Character Recognition Using Zoning Feature. Advances in
Electronics, Communication, and Computing. January 2018
13. Yadav, M., Mangal, D., Srinivasan, N., Ganzha, M.: Assamese Character Recognition using
Convolutional Neural Networks. May 2021
14. Dutta, P., Muppalaneni, N.: DigiNet Prediction of Assamese Handwritten Digits using Con-
volutional Neural Network. Concurrency and Computation: Practice and Experience. June
2021
15. Wikipedia, https://en.wikipedia.org/wiki/Assamese_alphabet, last accessed 2021/10/12