Text Type Extraction using a Deep Learning Solution at the Character Level
Md. Majedul Islam, A K M Shahariar Azad Rabby, Nazmul Hasan, Jebun Nahar and Fuad Rahman
Accepted to be presented at ICSCSP-2020: Soft Computing and Signal Processing, 21-22 August, 2020, Hyderabad, India
Description
Text-type detection of a document is an essential pre-processing step in the implementations of many document-processing solutions,
such as Optical Character Recognition (OCR) and machine translation. Specifically, text-type detection re-searches for Bangla is very rare,
with only a handful of solutions ever reported in the literature. In this paper, we present a lightweight, small footprint convolution-al neural network,
which detects handwritten and printed types of content directly from scanned mixed-type document images. The proposed model achieves 99.07% recognition
accuracy for this specific two-type classification problem.Text-type detection of a document is an essential pre-processing step in the implementations
of many document-processing solutions, such as Optical Character Recognition (OCR) and machine translation. Specifically, text-type detection re-searches
for Bangla is very rare, with only a handful of solutions ever reported in the literature. In this paper, we present a lightweight, small footprint
convolution-al neural network, which detects handwritten and printed types of content directly from scanned mixed-type document images. The proposed
model achieves 99.07% recognition accuracy for this specific two-type classification problem.
