Blind image deconvolution is an image processing technique that reconstructs a sharp, clear image from a blurred input when the blur kernel or point spread function is unknown. Unlike non-blind deconvolution, which operates with a known degradation model, blind deconvolution requires the simultaneous estimation of both the true latent image and the underlying blur process. Because infinitely many combinations of sharp images and blur operators can yield the same observed image, the problem is mathematically ill-posed and susceptible to noise amplification. To resolve this ambiguity, computational methods apply spatial or frequency domain constraints, statistical priors, optimization algorithms, or deep neural networks to accurately recover fine details and restore visual quality.
