An Efficient Kannada Handwritten Character Recognition Framework with Serial Dilated Cascade Network for Kannada Scripts

Abstract

The most significant problem present in the digitized world is handwritten character recognition and identification because it is helpful in various applications. The manual work needed for changing the handwritten character document into machine-readable texts is highly reduced by using the automatic identification approaches. Due to the factors of high variance in the writing styles beyond the globe, handwritten text size and low quality of handwritten text rather than printed text make handwritten character recognition to be very complex. The Kannada language has originated over the past 1000 years, where the consonants and vowels are symmetric in nature and also curvy, therefore, the recognition of Kannada characters online is very difficult. Thus, it is essential to overcome the above-mentioned complications presented in the classical Kannada handwritten character recognition model. The recognition of characters from Kannada Scripts is also difficult. Hence, this work aims to design a new Kannada handwritten character recognition framework using deep learning techniques from Kannada scripts. There are two steps to be followed in the proposed model that is collection of images and classification of handwritten characters. At first, essential handwritten Kannada characters are collected from the benchmark resources. Next, the acquired handwritten Kannada images are offered to the handwritten Kannada character recognition phase. Here, Kannada character recognition is performed using Serial Dilated Cascade Network (SDCN), which utilized the Visual Geometry Group 16 (VGG16) and Deep Temporal Convolution Network (DTCN) technique for the observation. When compared to the baseline recognition works, the proposed handwritten Kannada character recognition model achieves a significantly higher performance rate.