Handwritten Chinese Character Recognition Using Deep Neural Networks

Document Type

Oral Presentation

Campus where you would like to present

Ellensburg

Event Website

https://digitalcommons.cwu.edu/source

Start Date

16-5-2019

End Date

16-5-2019

Abstract

Chinese script is utilized by one-sixth of the world’s population. Unlike Latin, consisting only of twenty-six characters, the Chinese script comprises approximate ninety thousand characters. Due to the tremendous variation among these calligraphically rather complex characters, it is worthwhile for researchers to design and build fast and efficient technologies to classify handwritten Chinese characters. Recently, deep neural networks gained tremendous success in object recognition and different Latin based character recognition which led us to the idea to conduct several research attempts in this area by considering convolutional neural networks to recognize this complex script. To validate the efficiency and the robustness of the proposed Chinese character recognition system the CASIA HWDB1.1 database was considered. This dataset is a benchmark used by the research community to assess different handwriting recognizers. The dataset - built by 300 different writers - is considered a large one containing 1.1 million images split into 3755 most commonly used Chinese characters. Our various experiments culminated at 95% accuracy, which is a very promising result and comparable with the current state-of-the-art results.

Faculty Mentor(s)

Szilard Vajda

Department/Program

Computer Science

Jia_Handwritten Chinese Character Recognition Using Deep Neural Network.pdf (464 kB)
Slides for SOURCE 2019 presentation Song

Additional Files

Jia_Handwritten Chinese Character Recognition Using Deep Neural Network.pdf (464 kB)
Slides for SOURCE 2019 presentation Song

Share

COinS
 
May 16th, 12:00 AM May 16th, 12:00 AM

Handwritten Chinese Character Recognition Using Deep Neural Networks

Ellensburg

Chinese script is utilized by one-sixth of the world’s population. Unlike Latin, consisting only of twenty-six characters, the Chinese script comprises approximate ninety thousand characters. Due to the tremendous variation among these calligraphically rather complex characters, it is worthwhile for researchers to design and build fast and efficient technologies to classify handwritten Chinese characters. Recently, deep neural networks gained tremendous success in object recognition and different Latin based character recognition which led us to the idea to conduct several research attempts in this area by considering convolutional neural networks to recognize this complex script. To validate the efficiency and the robustness of the proposed Chinese character recognition system the CASIA HWDB1.1 database was considered. This dataset is a benchmark used by the research community to assess different handwriting recognizers. The dataset - built by 300 different writers - is considered a large one containing 1.1 million images split into 3755 most commonly used Chinese characters. Our various experiments culminated at 95% accuracy, which is a very promising result and comparable with the current state-of-the-art results.

https://digitalcommons.cwu.edu/source/2019/Oralpres/177