Degradation due to haziness, camera defocus and noise can be corrected using Image restoration. Only with the understanding of the deteriorating elements one can obtain the original image. Existing methods of image restoration have the limitations of suffering from bad convergence properties; the algorithms converging to local minima, and being unsuitable for real imaging applications. Few techniques, moreover, make constrictive presumptions on the PSF or the true image thereby limiting the algorithm's flexibility to different applications. Traditional approach involves de-blurring filters which are applied on the degraded images without the understanding of blur and its effectiveness. This paper is based on the approaches of AI that are applied for restoration problem in which images are distorted by a blur function and adulterated by some arbitrary noise. De-noising is enabled through the use of auto encoders while de-blurring is done through generative adversarial networks where a discriminator is used to analyze each output image given by the generator. The processing of satellite images is a major application of this proposed system of image restoration.

Keywords : Satellite Images, Generative Adversarial Networks, Autoencoders, Image Enhancement, De-Blur, DeNoise.