Ohio State nav bar

Seminar: Marilyn Vazquez - Non-Parametric Mixture Model Applied to Image Segmentation

Photo of Marilyn Vazquez
March 3, 2020
10:20AM - 11:15AM
MBI Auditorium, Jennings Hall 355

Date Range
Add to Calendar 2020-03-03 10:20:00 2020-03-03 11:15:00 Seminar: Marilyn Vazquez - Non-Parametric Mixture Model Applied to Image Segmentation Marilyn Vazquez Postdoctoral Fellow, MBI Mixture models have many uses in data science, including clustering and image segmentation. We present an unsupervised image segmentation method applied to natural images. In our research, we model an image as a mixture of color distributions, each coming from a distinct segment in the image. The goal is to estimate the color distributions from all the regions in the image and use them to segment the image. To solve for the distributions, we developed a scheme that solves a subset of the equations in our model, uses least squares to find the missing entries, and then performs coordinate descent to refine the approximation. The parameters in our model are estimated by measuring how well they recover the observed data. We present results on natural images and materials. This talk is free and open to the public. MBI Auditorium, Jennings Hall 355 Mathematical Biosciences Institute mbi-webmaster@osu.edu America/New_York public

Marilyn Vazquez

Postdoctoral Fellow, MBI


Mixture models have many uses in data science, including clustering and image segmentation. We present an unsupervised image segmentation method applied to natural images. In our research, we model an image as a mixture of color distributions, each coming from a distinct segment in the image. The goal is to estimate the color distributions from all the regions in the image and use them to segment the image. To solve for the distributions, we developed a scheme that solves a subset of the equations in our model, uses least squares to find the missing entries, and then performs coordinate descent to refine the approximation. The parameters in our model are estimated by measuring how well they recover the observed data. We present results on natural images and materials.

This talk is free and open to the public.

 

 

Events Filters: