Optical Character Recognition System for Czech Language Using Hierarchical Deep Learning Networks

Abstract

Optical character recognition (OCR) systems play vital role in pattern recognition research. With rapid growth of OCRs for different languages developing OCR for Czech language is looked upon as positive aspect for people speaking Czech language. In this paper, we develop OCR system for Czech language using hierarchical fuzzy convolutional neural networks (HFCNN). We present end-to-end framework that includes pre-processing activities, segments text image, classifies characters and performs recognition. The feature extraction is performed through fuzzy Hough transform. The feature based classification is performed through HFCNN. A comprehensive assessment of proposed method is performed through publicly available Czech language dataset. OCR recognition accuracy is a major concern. There is always an inherent degree of vagueness and impreciseness present in reallife data. Due to this recognition system is treated here through fuzzy sets encompassing indeterminate uncertainty. The simulation studies reveal that deep learning based OCR for Czech language performs consistently better than traditional models. The experimental results demonstrate efficiency of proposed approach.