EE Seminar: Co- Occurring Clusters Denoising
Speaker: Yair Shefi
M.Sc. student under the supervision of Prof. Shai Avidan and Yacov Hel-Or
Wednesday, October 24th 2018 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
Co- Occurring Clusters Denoising
Abstract
We suggest a method called Co- Occurring Context (CoC) denoising which applies a new prior to natural images, sparse labels Co-Occurrence matrix.
We suggest achieving high image restoration by decreasing the elements count in the
Co-Occurrence matrix. Our Co-Occurrence matrix is a statistical representation of the pixels labels created by a clustering algorithm, exposing the statistical connection of two labels been assigned in nearby spatial window. We implement context clustering by using this prior to regularize our clustering objective function.
Context clustering prefer labels that can blend in their spatial neighborhood hence preserve context. Our denoising method takes advantage of the sub-problem and creates a restoration to each cluster separately using simple yet very effective method.
