The feature image associated with a model

We show feature images associated with models.

In the experiment of Corpus callosum border segmentation, the model C=(tx, ty, theta, s, b1, b2, b3, b4, b5, b6), where tx and ty is the translation, theta is the rotation, s is the scale, b1...b6 are the shape parameters.

We assume the initial error range of the pose is [20 20 pi/9 0.2]. The initial error range of the shape parameters is 3*sqrt(lambda). Please refer to section 5.1 in the paper for more details.

The following subimages x(I,C)'s are extracted from an image I by varying a chosen parameter of the model C in the feasible region and setting the remaining parameters to the ground truth. 

Because the global scale variation is small, we fix the scale of subimages. The fitness of the scale is measured by the position of the subimages corresponding to the control points. We only show three control points here. In the experiments. 32 control points are used. Please refer to section 3.1 in the paper for more details.

First dimension

(click the image to see the video)

Second dimension

(click the image to see the video)

Third dimension

(click the image to see the video)

Forth dimension

(click the image to see the video)

Fifth dimension

(click the image to see the video)

Sixth dimension

(click the image to see the video)

Seven dimension

(click the image to see the video)

Eight dimension

(click the image to see the video)

Nineth dimension

(click the image to see the video)

Tenth dimension

(click the image to see the video)