Principal Component 2-D Long Short-Term Memory for Font Recognition on Single Chinese Characters

Area of Science:

• Computer Science
• Artificial Intelligence
• Pattern Recognition

Background:

• Intelligent applications increasingly rely on Optical Character Recognition (OCR), driving demand for accurate Chinese Character Font Recognition (CCFR).
• Traditional CCFR systems struggle with noisy data, limiting their practical application.
• Recognizing Chinese character fonts is crucial for advancing OCR technology.

Purpose of the Study:

• To develop a robust Chinese character font recognition system capable of handling noisy data.
• To address the limitations of traditional CCFR methods in real-world scenarios.
• To propose a novel deep learning approach for sequence classification in CCFR.

Main Methods:

• Character font recognition framed as a sequence classification problem.
• Development of the Principal Component 2DLSTM (PC-2DLSTM) algorithm.
• Integration of a principal component convolution layer with 2-D Long Short-Term Memory (2DLSTM) for noise reduction and contextual processing.

Main Results:

• The PC-2DLSTM algorithm effectively removes noise and extracts complete font information.
• 2DLSTM component captures long-range contextual information, enhancing character trajectory analysis.
• Experimental results demonstrate the superiority of PC-2DLSTM over existing state-of-the-art CCFR methods on a standard dataset.

Conclusions:

• PC-2DLSTM offers a robust and effective solution for Chinese character font recognition, particularly in noisy conditions.
• The proposed method significantly improves recognition accuracy compared to traditional approaches.
• This research contributes to the advancement of intelligent OCR applications.