Image restoration involves methods to recover a clean, undistorted image from a degraded version. Unlike general enhancement which may not have a specific model of degradation, restoration assumes a mathematical model for how the image got degraded and tries to invert it. Examples include deblurring (if an image is blurred by camera shake or defocus, attempt to sharpen it by deconvolution), denoising (remove noise by modeling it as, say, Gaussian additive noise), inpainting (fill in missing or corrupted parts of an image in a plausible way), colorization (restore color to black-and-white images), and super-resolution (increase the resolution). Restoration algorithms may use techniques from linear filtering, optimization, or deep learning (e.g., convolutional networks trained to map noisy to clean images). The measure of success is often fidelity to the original (unknown) image – PSNR (peak signal-to-noise ratio) and SSIM (structural similarity) are common quantitative metrics for restoration quality.