We addresses one of the key problems of image processing, which is restoration of a latent image from its degraded observations. As the main source of degradation we consider convolution with a unknown blur. The restoration task is a blind deconvolution problem, which is intrinsically ill-posed. We will study conditions under which the single-channel (single observation) case can be expressed as multichannel (multiple observations) blind deconvolution, which is a better posed problem. We will use state-of-the-art sampling methods to estimate full posterior distributions and infer the unknown blur and latent image from the distribution. Practical scenarios require more complex convolution models and they will be appropriately tackled. In photography, complex camera-object motion results in space-variant blind deconvolution. In analysis of dynamic medical imaging, blind source separation encapsulates the deconvolution problem. In confocal microscopy, we face deconvolution in 3D. Implementation on different platforms including hand-held devices will be carried out.