Blind text images deblurring based on a generative adversarial network

Abstract

Recently, text images deblurring has achieved advanced development. Unlike previous methods based on hand‐crafted priors or assume specific kernel, the authors recognise the text deblurring problem as a semantic generation task, which can be achieved by a generative adversarial network. The structure is an essential property of text images; thus, they propose a structural loss function and a detailed loss function to regularise the recovery of text images. Furthermore, they learn from the coarse‐to‐fine strategy and present a multi‐scale generator, which is utilised for sharpening the generated text images. The model has a robust capability of generating realistic latent images with photo‐quality effect. Extensive experiments on the synthetic and real‐world blurry images have shown that the proposed network is comparable to the state‐of‐the‐art methods.

High‐speed target tracking algorithm for the pulse‐sequence‐based image sensor

“…Qing Qi et al achieved blind text images deblurring based on a generative adversarial network. The experimental results showed that the proposed network was comparable to the state‐of‐the‐art methods [30]. These are image deblurring methods based on the traditional image sensors.…”

Taming mode collapse in generative adversarial networks using cooperative realness discriminators

“…By controlling the angles ∠(w r ∇v r , ∇v m D ) and ∠(w f ∇v f , ∇v m D ) to be acute (see Figure 2) when one of them becomes obtuse, one can prevent the issue that training may benefit one loss but significantly harm the other. However, in previous models [4][5][6][7][8][9][30][31][32][33][34], the loss function of the discriminator consists of two equally weighted parts, that is,…”

“…Since the introduction of GANs, many variants have been proposed [2][3][4][5] to improve the generated image quality. Although GANs have already been successfully used in image generation and image editing [6][7][8][9], they are notoriously difficult to train, and it has been observed that they often suffer from mode collapse [10,11], which causes the generator network to produce results with poor generalizability; that is, a limited variety of samples are produced, but many other modes are missed.…”