# US 2013/0279801 A1 · Method and apparatus for image segmentation

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2013/0279801 A1
- Document ID: US20130279801A1
- Status: Published application
- Kind: A1
- Application number: 13/862,413
- Filed: 2013-04-13
- Published: 2013-10-24
- Earliest claimed: 2012-04-20
- Assignees: Thomson Licensing
- Applicants: Thomson Licensing
- Inventors: Bjoern Scheuermann; Oliver Jakob Arndt; Bodo Rosenhahn
- Web page: https://patentgrep.com/patent/US20130279801A1

## CPC

- G06T 7/11: Region-based segmentation
- G06T 7/143: involving probabilistic approaches, e.g. Markov random field [MRF] modelling
- G06T 7/187: involving region growing; involving region merging; involving connected component labelling
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL

## Abstract

A method and an apparatus for multi-label segmentation of an image are described. In a first step user defined labels are determined for one or more pixels of the image. Then a fraction of the pixels of the image for which no user defined label is determined is pre-initialized. Finally, a cellular automaton-based segmentation of the image is performed using the user defined labels and the pre-initialized pixels.

## Claims

### Claim 1

A method for segmentation of an image, the method comprising the steps of:
determining user defined labels for one or more pixels of the image;
pre-initializing a fraction of the pixels of the image for which no user defined label is determined; and
performing a cellular automaton-based segmentation of the to image using the user defined labels and the pre-initialized pixels.

### Claim 2 (depends on 1)

The method according to claim 1, wherein the fraction of pixels which are pre-initialized is user-settable.

### Claim 3 (depends on 1)

The method according to claim 1, wherein a strength assigned to the pixels for which a user defined label is determined is larger than a strength assigned to the pre-initialized pixels.

### Claim 4 (depends on 1)

The method according to claim 1, wherein Gaussian mixture models are used to determine labels for the pre-initialized pixels.

### Claim 5 (depends on 1)

The method according to claim 1, wherein the step of pre-initializing a fraction of the pixels of the image for which no user defined label is determined comprises:
assigning label probabilities of the pixels to n-dimensional vectors, where n is the number of labels;
performing an n-sphere transformation of the n-dimensional vectors;
smoothing each dimension of the n-sphere coordinates to obtain a smoothed vector field; and
retransforming the smoothed vector field.

### Claim 6 (depends on 1)

The method according to claim 1, wherein a weighting function is applied to a strength of a pixel, which maintains the strength for small color differences between neighboring pixels and decreases the strength for larger color differences.

### Claim 7 (depends on 1)

The method according to claim 1, wherein the cellular automaton-based segmentation uses a GrowCut algorithm.

### Claim 8 (depends on 1)

The method according to claim 1, wherein the segmentation is a binary or multi-label segmentation.

### Claim 9

An apparatus for segmentation of an image, wherein the apparatus is adapted to perform the steps of:
determining user defined labels for one or more pixels of the image;
pre-initializing a fraction of the pixels of the image for which no user defined label is determined; and
performing a cellular automaton-based segmentation of the image using the user defined labels and the pre-initialized pixels.
