Paper

Summary

論文簡介/

Contribution:

The adversarial training is used to automatically extract features of LN OCT (視網膜光學斷層掃描) images.
The frequency loss is first added to the loss function to suppress the noise in the HN region and preserve the structural.
A new denoising framework, DN-GAN.
Speckle noise (Multiplicative noise)
Image pixel value + random value = AWGN
Image pixel value * random value = Speckle noise
Cause by the receiver from the radar, CT, ultrasound, the different reflection from the rough surface or different tissue.