Machine Learning For Image Restoration: A Review Of Methods, Trends, And Challenges

Image restoration the task of recovering degraded or damaged images has become essential across many technical domains, including space imaging, medical imaging, and several post-processing applications. Most restoration techniques begin by modeling the degradation process that corrupts an image, typically involving blur and noise, and then attempt to reconstruct an approximation of the original image. However, in real-world scenarios, degradation is often unknown, requiring the simultaneous estimation of both the true image and the blurring function directly from the observed degraded image, without relying on prior knowledge of the blur mechanism. This thesis proposes a novel digital image restoration approach based on punctual kriging, supported by multiple machine learning algorithms. The work focuses on restoring images corrupted by Gaussian noise by achieving an effective trade-off between two competing goals: producing smooth, visually pleasing results while preserving edge details and structural integrity.