Digging Deeper into CRNN Model in Chinese Text Images Recognition

TL;DR

This paper enhances Chinese text image recognition by extending CRNN to multi-row images, introducing Line-DDeCAE for line recovery, and applying knowledge distillation for model compression, validated through experiments on artificially generated data.

Contribution

It proposes methods to extend CRNN for multi-row images, recover box lines with Line-DDeCAE, and compress models using knowledge distillation, addressing limitations in recognizing complex Chinese text images.

Findings

Extended CRNN to multi-row images successfully.

Line-DDeCAE effectively recovers box lines in images.

Knowledge distillation reduces model size without accuracy loss.

Abstract

Automatic text image recognition is a prevalent application in computer vision field. One efficient way is use Convolutional Recurrent Neural Network(CRNN) to accomplish task in an end-to-end(End2End) fashion. However, CRNN notoriously fails to detect multi-row images and excel-like images. In this paper, we present one alternative to first recognize single-row images, then extend the same architecture to recognize multi-row images with proposed multiple methods. To recognize excel-like images containing box lines, we propose Line-Deep Denoising Convolutional AutoEncoder(Line-DDeCAE) to recover box lines. Finally, we present one Knowledge Distillation(KD) method to compress original CRNN model without loss of generality. To carry out experiments, we first generate artificial samples from one Chinese novel book, then conduct various experiments to verify our methods.