Unsupervised Structure-Texture Separation Network for Oracle Character Recognition

TL;DR

This paper introduces an unsupervised learning framework called STSN that effectively separates structure and texture features in oracle character images, enabling improved recognition across different data domains without requiring labeled scanned data.

Contribution

The novel STSN model performs joint disentanglement, domain adaptation, and recognition for oracle characters, addressing the challenge of scarce labeled scanned data.

Findings

STSN outperforms existing adaptation methods on Oracle-241 dataset.

It improves recognition accuracy of scanned oracle characters, even with noise and damage.

The model effectively disentangles structure and texture features for better domain transfer.

Abstract

Oracle bone script is the earliest-known Chinese writing system of the Shang dynasty and is precious to archeology and philology. However, real-world scanned oracle data are rare and few experts are available for annotation which make the automatic recognition of scanned oracle characters become a challenging task. Therefore, we aim to explore unsupervised domain adaptation to transfer knowledge from handprinted oracle data, which are easy to acquire, to scanned domain. We propose a structure-texture separation network (STSN), which is an end-to-end learning framework for joint disentanglement, transformation, adaptation and recognition. First, STSN disentangles features into structure (glyph) and texture (noise) components by generative models, and then aligns handprinted and scanned data in structure feature space such that the negative influence caused by serious noises can be avoided when adapting. Second, transformation is achieved via swapping the learned textures across domains and a classifier for final classification is trained to predict the labels of the transformed scanned characters. This not only guarantees the absolute separation, but also enhances the discriminative ability of the learned features. Extensive experiments on Oracle-241 dataset show that STSN outperforms other adaptation methods and successfully improves recognition performance on scanned data even when they are contaminated by long burial and careless excavation.