一种用于甲骨字符提取的双分支融合网络

A dual-branch fusion network for extracting Oracle characters

为了从甲骨拓片图像中自动提取甲骨字符信息,本文基于深度神经网络构建了一个甲骨字符提取的双分支融合网络(dual-branch fusion network for extracting Oracle characters,EOCNet).EOCNet包含3个基本特点:首先,为了能够利用生成网络较强的结构信息描述能力,EOCNet以对抗生成网络(generative adversarial network,GAN)为基本骨架,将甲骨字符提取问题视为图像到图像的转换任务;其次,为了能利用语义分割网络较强的拓片图像背景和甲骨字符的区分能力,EOCNet将语义分割网络融入生成器网络,并通过空间注意力模型(spatial attention module,SAM)来提高甲骨字符区域特征在生成甲骨字符图像中的作用;再次,为获取内容完整且细节清晰的生成结果,EOCNet结合全局判别器网络和局部判别器网络对生成的甲骨字符图像进行一致性判别.实验结果表明,相比于主流的基于深度学习的图像生成和分割方法,本文模型能够生成更高质量的甲骨字符图像.

To automatically extract Oracle bone inscriptions(OBI)information from Oracle bone rubbing images,we propose a dual-branch fusion network for extracting Oracle characters(EOCNet)based on deep neural networks.Specifically,EOCNet includes three basic characteristics.First,for the purpose of utilizing the powerful ability of generative networks to describe structural information,EOCNet takes generative adversarial network(GAN)as its backbone,regarding the task of OBI character extraction as a kind of image-to-image conversion.Second,for the sake of employing the strong capacity of semantic segmentation networks to distinguish the background and OBI character in rubbing images,EOCNet integrates a semantic segmentation network into its generator part,and improves the role of OBI characters in the generation of OBI character images through a spatial attention module(SAM).Third,for obtaining a complete and detailed result,EOCNet combines a global discriminator network and a local discriminator network to judge the consistency of the generated OBI character image.Experimental results show that,compared with the state-of-the-art image generation methods and semantic segmentation methods based on deep learning,the proposed model can generate higher quality OBI character images.