Abstract:
In microscopy, eliminating blurriness and obtaining a sharp and accurate edge of observed objects in multiple blurriness is a key issue. Herein, a deblurring model called...Show MoreMetadata
Abstract:
In microscopy, eliminating blurriness and obtaining a sharp and accurate edge of observed objects in multiple blurriness is a key issue. Herein, a deblurring model called De-super-resolution generative adversarial network (De-SRGAN) is proposed, in which multiple discriminators are set. The multidiscriminator architecture can provide stable training, which helps achieve effective deblurring results in multiple blurriness. Moreover, considering the property of the edge in the frequency domain, a perceptual loss term in the frequency domain is constructed as an additional factor. Results show that De-SRGAN can generate deblurred images with low mean squared error (mse)/mean absolute error (MAE), high peak signal-to-noise ratio (PSNR)/structural similarity (SSIM), high perceptual quality, and high accuracy. This study contributes to improving the accuracy of size measurement due to the deblurred images with sharp and accurate edges.
Published in: IEEE Transactions on Instrumentation and Measurement ( Volume: 73)