BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Divide and Conquer: Patch-based Image Denoising\, Restoration\, an d Beyond - Mario Figueiredo (Universidade de Lisboa\; Instituto Superior T écnico\, Lisboa) DTSTART:20171102T140000Z DTEND:20171102T145000Z UID:TALK94360@talks.cam.ac.uk CONTACT:INI IT DESCRIPTION:Patch-based image processing methods can be seen as an applica tion of the &ldquo\;divide and conquer&rdquo\; strategy: since it is admit tedly too difficult to formulate a global prior for an entire image\, meth ods in this class process overlapping patches thereof\, and combine the re sults to obtain an image estimate. A particular class of patch-based metho ds uses Gaussian mixtures models (GMM) to model the patches\, in what can be seen as yet another application of the divide and conquer principle\, n ow in the space of patch configurations. Different components of the GMM s pecialize in modeling different types of typical patch configurations. Alt hough many other statistical image models exist\, using a GMM for patches has several relevant advantages: (i) the corresponding minimum mean square d error (MMSE) estimate can be obtained in closed form\; (ii) the variance of the estimate can also be computed\, providing a principled way to weig ht the estimates when combining the patch estimates to obtain the full ima ge estimate\; (iii) the GMM parameters can be estimated from a dataset of clean patches\, from the noisy image itself\, or from a combination of th e two\; (iv) theoretically\, a GMM can approximate arbitrarily well any pr obability density (under mild conditions). In this talk\, I will overview the class of patch/GMM-based approaches to image restoration. After review ing the first members of this family of methods\, which simply addressed d enoising\, I will describe several more recent advances\, namely: use of c lass-adapted GMMs (i.e.\, tailored to specific image classes\, such as fac es\, fingerprints\, text)\; tackling inverse problems other than denoising (namely\, deblurring\, hyperspectral super-resolution\, compressive imagi ng)\, by plugging GMM-based denoisers in the loop of an iterative algorith m (in what has recently been called the plug-and-play approach)\; joint re storation/segmentation of images\; application to blind deblurring. This i s joint work with Afonso Teodoro\, José\; Bioucas-Dias\, Marina Ljub enovi\;ć\;\, and Milad Niknejad. LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR