Unsupervised Image Segmentation Based on a Background Likelihood Estimation
AssigneeIntel Corporation
InventorsTzachi Hershkovich, Yevgeny Priziment, Itay Benou
Life of this patent
Estimating the term…
Abstract
A method for unsupervised image segmentation based on a background likelihood estimation is described herein. The method includes calculating a distance for each pixel in an input image and transforming each distance into a probability of the distance of belonging to a background of an input image or a foreground of the input image. The method also includes transforming each probability into a log likelihood value to form a unary term and minimizing an energy function including the unary term to obtain a set of foreground labels and a set of background labels.
Claims25
A method for unsupervised image segmentation based on a background likelihood estimation, comprising:
- calculating a distance for each pixel in an input image;
- calculating a probability of each pixel in the input image as belonging to a background of the input image or a foreground of the input image based on the distance;
- transforming each probability into a log likelihood value to form a unary term;
- and
- minimizing an energy function including the unary term to obtain a set of foreground labels and a set of background labels.
A system for unsupervised image segmentation based on a background likelihood estimation, comprising:
- a distance calculator to calculate a distance for each pixel in an input image;
- a probability calculator to calculate a probability of each pixel in the input image as belonging to a background of the input image or a foreground of the input image based on the distance;
- a likelihood calculator to transform each probability into a log likelihood value and form a unary term;
- and
- a function minimizer to minimize an energy function including the unary term to obtain a set of foreground labels and a set of background labels.
At least one non-transitory computer-readable medium, comprising instructions to direct a processor to:
- calculate a distance for each pixel in an input image;
- calculate a probability of each pixel in the input image as belonging to a background of the input image or a foreground of the input image based on the distance;
- transform each probability into a log likelihood value to form a unary term;
- and
- minimize an energy function including the unary term to obtain a set of foreground labels and a set of background labels.
Dependencies are inferred from the claim text.