US 2015/0030219 A1Published application

Method and apparatus for shape based deformable segmentation of multiple overlapping objects

AssigneeRutgers, The State University of New Jersey

InventorsAnant Madabhushi, Sahirzeehan Ali

Life of this patent

Estimating the term…

Loading patent…

Abstract

The present invention provides a system and method for simultaneous variational and adaptive segmentation of single non-overlapping and multiple overlapping/occluded-objects in an image scene. The method according to the present invention synergistically combines shape priors with boundary and region-based active contours in a level set formulation with a watershed scheme for model initialization for identifying and resolving multiple object overlaps in an image scene. The method according to the present invention comprises learning a shape prior for a target object of interest and applying the shape prior into a hybrid active contour model to segment overlapping and non-overlapping objects simultaneously.

Claims45

Dependent claims

A method for segmenting single non-overlapping objects or multiple overlapping/occluded objects in an image scene via a hybrid adaptive active contour model incorporating shape priors, comprises:

(a) inputting the image scene via a computing device;
(b) collecting a plurality of shapes via the computing device, wherein the shapes correspond to the objects in the image scene;
(c) determining a set of alignment parameters via the computing device, by performing a similarity transformation;
(d) aligning the plurality of collected shapes in (b) via the computing device to create a plurality of aligned shapes, such that the plurality of aligned shapes have the same pose compared to a reference shape based on the similarity transformation of (b);
(e) creating a training set via the computing device, wherein the training set comprises the plurality of aligned shapes in (d), wherein each aligned shape of the plurality of aligned shapes is embedded as a zero level set of a higher dimensional surface;
(f) generating a shape prior via the computing device;
(g) integrating a shape term via the computing device, wherein each aligned shape of the plurality of aligned shapes is embedded as the zero level set of the higher dimensional surface, and wherein the shape term is a signed distance function (SDF) used to encode the distance of the shape contour of each aligned shape to the nearest point in the shape space of the higher dimensional surface;
(h) integrating a boundary-based term via the computing device to direct the hybrid adaptive active contour model to a boundary of at least one object of the objects in the image scene;
(i) integrating a region homogeneity term via the computing device, wherein the region homogeneity term drives the shape term towards a homogeneous intensity region using the shape prior;
(j) initializing the hybrid adaptive active contour model via the computing device by a watershed transformation method comprising delineating object boundaries of at least one target object in the image scene;
(k) performing contour propagation of the delineated shape boundaries via the computing device by evolving the integrated synergistic shape functional by minimizing variational energy;
and
(l) achieving final segmentation of the image scene via the computing device.

10 more elements

A system for segmenting single non-overlapping objects or multiple overlapping/occluded objects in an image scene via a hybrid adaptive active contour model incorporating shape priors, comprises:

(a) a computing device configured to input the image scene;
(b) the computing device configured to collect a plurality of shapes, wherein the shapes correspond to the objects in the image scene;
(c) the computing device configured to determine a set of alignment parameters by performing a similarity transformation;
(d) the computing device configured to align the plurality of collected shapes in (b) to create a plurality of aligned shapes, such that the plurality of aligned shapes have the same pose compared to a reference shape based on the similarity transformation of (b);
(e) the computing device configured to create a training set, wherein the training set comprises the plurality of aligned shapes in (d), wherein each aligned shape of the plurality of aligned shapes is embedded as a zero level set of a higher dimensional embedding function describing a surface;
(f) the computing device configured to generate a shape prior;
(g) the computing device configured to integrate a shape term, wherein each aligned shape of the plurality of aligned shapes is embedded as the zero level set of the higher dimensional surface, and wherein the shape term is a signed distance function (SDF) used to encode the distance of the shape contour of each aligned shape to the nearest point in the shape space of the higher dimensional surface;
(h) the computing device configured to integrate a boundary-based term to direct the hybrid adaptive active contour model to a boundary of at least one object of the objects in the image scene;
(i) the computing device configured to integrate a region homogeneity term, wherein the region homogeneity term drives the shape term towards a homogeneous intensity region using the shape prior;
(j) the computing device configured to initialize the hybrid adaptive active contour model by a watershed transformation method comprising delineating object boundaries of at least one target object in the image scene,
(k) the computing device configured to perform contour propagation of the delineated shape boundaries by evolving the integrated synergistic shape functional by minimizing variational energy;
and
(l) the computing device configured to achieve final segmentation of the image scene.

10 more elements

Dependencies are inferred from the claim text.

Facts

Application: 13/978,763Filed 10 Jan 2012
Priority: 10 Jan 2011
Published: 29 Jan 2015
Term ends: Not estimatedThe source records don’t support an estimate.

Classification12 codes

G06T 7/12
Segmentation; Edge detection: edge-based segmentation
G06T 7/149
Segmentation; Edge detection: involving deformable models, e.g. active contour models
G06V 10/26
Image preprocessing: segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g.…
G06V 20/695
Microscopic objects, e.g. biological cells or cellular parts: preprocessing, e.g. image segmentation