Isolated bangla handwritten character & digit recognition using convolutional neural network
Abstract
This paper proposes a mechanism of Handwritten Letter and Digit Recognition (HLDR) to
decipher images of Bangla handwritten characters into electronically editable format, which
holds an important role in augmenting and digitalizing many analog application, which will
not only paves the way to further research but also have many practical applications in
current times.The mechanisms of HLDR has been studied broadly in the last half century,
moreover, the rapid growth of computational power and main memory breaks the barrier and
gives the opportunity for the implementation of more efficient and complex HLDR
methodologies, which creates an increasing demand on many forthcoming application
domains. In the field of pattern recognition one of the most productive way of achieving
higher accuracy or lower error rate is to adopt an architecture that is deep, optimized and can
process a large number of data. Therefore, this paper propose that using deeper residual
network [1](ResNet) architecture and recently released Bangla-lekha dataset [2], we can
achieve a result which is higher than any research that has been done before.