Image Inpainting Technique Based on Smart Terminal: A Case Study in CPS Ancient Image Data

Abstract

Cyber-physical system (CPS) can intelligently feel the interactive information between data and terminal. The combination of CPS data and intelligent terminal can make terminal devices communicate with each other, coordinate operation, and share information in a complex environment. At present, the demand for smart terminal technology in university libraries and national cultural museums is also increasing. However, the data for smart terminal display and development of shared services face the problem that the original data is difficult to collect and the incomplete images are difficult to repair. In order to display the ancient books in a smart terminal, we have established a model for image inpainting through deep learning. In this paper, we set up a smart terminal display solution to display the digital protection effect of incomplete character images. Then, we propose a model based on Conditional Generative Adversarial Network (CGAN) to solve the problem of repairing incomplete character images. We put this model into the computing module and deployed it to the cloud computing platform of the smart terminal. Moreover, we construct a handwritten Yi character image data set in the cloud. To verify the CGAN calculation model, we design and construct two quality detection tests. According to the experimental results, the image inpainting quality of the CGAN is better than the traditional image restoration technology. Based on the experimental analysis and evaluation criteria, we find that the CGAN calculation model is of great significance for ancient book inpainting.INDEX TERMS CPS data, smart terminal, character image, conditional generative adversarial network.

An Auto-Encoder of Inscription Character Inpainting based on Branch Convolutional Channel Attention Module

“…Classical models based on GANs include the terminal-based intelligent image restoration technique proposed by Yu et al and a dual discriminator GAN for ancient Yi characters inpainting proposed by Chen et al [6,7]. The terminal-based intelligent image inpainting technique transforms the characters to be restored into one-dimensional vectors as conditional inputs into Conditional Generative Adversarial Networks (CGAN) for character generation [8].The dual discriminator GAN method introduces a character filter based on the Deep Convolutional Generative Adversarial Networks (DCGAN) to screen the noise and then uses the screened noise to generate characters to complete the Yi character inpainting task [9].…”

An Auto-Encoding Network for Binary Inscription Character Inpainting based on U-Net and Self-Attention

“…Most of the character inpainting still takes digital preservation of ancient documents as the main application. Among them, Yu et al and Chen et al proposed an image inpainting technique based on a smart terminal and a dual discriminator GAN for ancient Yi characters inpainting to repair Yi characters, respectively [6,11]. The smart terminal image inpainting technique transforms the characters to be restored into one-dimensional vectors and inputs them as conditions into Conditional Generative Adversarial Networks (CGAN) for character generation [12].…”

Binary Inscription Character Inpainting Based on Improved Context Encoders

“…Yu et al improve CGAN(Conditional Generative Adversarial Networks) by proposing an intelligent terminal-based image inpainting technique to restore Yi characters [21]. The Yi character to be restored is flattened into a one-dimensional vector as the condition and random noise input to the generator, allowing the generator to generate the specified Yi character according to the label to achieve Yi character inpainting [22]. The Yi character from literature [22] is shown in Figure 1.…”

“…The Yi character to be restored is flattened into a one-dimensional vector as the condition and random noise input to the generator, allowing the generator to generate the specified Yi character according to the label to achieve Yi character inpainting [22]. The Yi character from literature [22] is shown in Figure 1. Chen et al improve DCGAN(Deep Convolutional Adversarial Networks) and propose a character inpainting model based on the dual discriminator GANs, adding a character filter based on DCGAN [23].…”