org.openimaj.image.model.landmark

Interface LandmarkModel<I extends Image<?,I>>

Type Parameters:

I - Type of image

All Known Implementing Classes:

FNormalLandmarkModel, FPatchLandmarkModel

public interface LandmarkModel<I extends Image<?,I>>

A LandmarkModel models local image content and provides functionality to move a point in an image to a nearby point with a lower cost than at the initial point.

Landmarks are normally associated with points in PointLists and thus quite often have an associated "intrinsic scale". This scale can be used to dynamically change the size of the support region of a landmark so that is scales with the PointLists intrinsic scale.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Method Summary

Modifier and Type	Method and Description
float	computeCost(I image, Point2d point, PointList pointList) Evaluate the cost function using the given image and point.
void	updateModel(I image, Point2d point, PointList pointList) Update the internal model of local image content by adding information from the provided image.
ObjectFloatPair<Point2d>	updatePosition(I image, Point2d initial, PointList pointList) Estimate an improved fit based on a local neighbourhood search.

Method Detail

updateModel

void updateModel(I image,
                 Point2d point,
                 PointList pointList)

Update the internal model of local image content by adding information from the provided image.

Parameters:

image - the image

point - the point in the image representing the landmark

pointList - the pointList to which the point belongs

computeCost

float computeCost(I image,
                  Point2d point,
                  PointList pointList)

Evaluate the cost function using the given image and point. Lower costs indicate better fits.

Parameters:

image - the image

point - the point in the image

pointList - the pointList to which the point belongs

Returns:

the cost

updatePosition

ObjectFloatPair<Point2d> updatePosition(I image,
                                        Point2d initial,
                                        PointList pointList)

Estimate an improved fit based on a local neighbourhood search. Returns the new best point and the distance moved normalised by the size of the search area. If the point didn't move, then the distance would be 0; if the point moved to the extremity of the search region, it would be 1.0.

Parameters:

image - the image

initial - the initial point in the image

pointList - the pointList to which the point belongs

Returns:

the updated point