A Novel Unified Scheme for Missing Data Suggestion based on Collaborative Generative Adversarial Network

EasyChair Preprint 5243

Abstract

Image processing is generally a complicated task if it is performed for the medical images, as these image files have different types of attributes that has different properties. There are often times when these attributes are not measured properly and sometimes the medical imaging process produces an image with distorted or diminished pixels in some part of the image, and this will create a problem while image processing is performed. To solve this, we have introduced a model which works on Recurrent Neural Networks (RNN) to predict the possible distorted or the missing parts in the image and then using the Generative Adversarial Network (GAN) that uses Convolutional Neural Network (CNN) in the generator to fix the missing pixel in the image. The discriminator part of the GAN is trained to trace the error made by the generator which is recorded as the penalty in the discriminator and is sent as a feedback to the generator to fix the error and produce an acceptable image. The process is like the Minimax game played between the generator and the discriminator that minimize the error at every time the image is generated. This error fixing part of the GAN proves very useful in order to create a image of the required resolution without compromising with the other sensitive parameters present in the image.

Keyphrases: Gaussian Mixture Model, Generative Adversarial Network, Long Short-Term Memory, Recurrent Neural Network, simple recurrent network

@booklet{EasyChair:5243,
  author    = {P Sabitha and Anubhav Saha and Yateendra Mishra and Anirban Pal},
  title     = {A Novel Unified Scheme for Missing Data Suggestion based on Collaborative Generative Adversarial Network},
  howpublished = {EasyChair Preprint 5243},
  year      = {EasyChair, 2021}}